SPB Git

Repository Public

conversation.json 187.4 KB 1 lines json

1{"sessionId":"3d04acf2-235e-46e3-8c2d-bf26b0ba3c81","title":"J ai téléversé data/us_macro_quarterly.csv (macro trimestrie...","messages":[{"question":"J ai téléversé data/us_macro_quarterly.csv (macro trimestrielle États-Unis, 1959T1–2009T3 : PIB réel, consommation, investissement, IPC, M1, taux des bons du Trésor 3 mois, chômage, inflation, taux réel). Question de recherche : la relation entre inflation, chômage et taux court (courbe de Phillips et règle de politique monétaire) est-elle stable sur la période, ou y a-t-il des changements de régime, et quel modèle prévoit le mieux l inflation hors échantillon ? Mène l étude de bout en bout : explore les données ; teste la stationnarité ; détecte les ruptures structurelles avec une bibliothèque spécialisée de détection de points de rupture (elle n est pas dans l environnement de base : installe ce qu il te faut) ; estime et compare au moins trois modèles de prévision de l inflation (par exemple AR de référence, VAR, modèle à changement de régime markovien ou autre que tu juges pertinent) avec une évaluation hors échantillon en fenêtre glissante ou extensible et un test de comparaison de précision prédictive ; fais les diagnostics des résidus ; vérifie la robustesse des conclusions. Lance les estimations lourdes en parallèle comme programmes du workspace. Tiens le registre des expériences, y compris ce qui échoue. Livre un rapport de recherche avec des figures de qualité publication.","answer":"\n\nJe lance le MS en fenêtre 40 pour la robustesse (en parallèle), puis j'évalue.\n\nJe consigne les expériences restantes (y compris les échecs) et la mémoire de recherche.\n\n# Inflation, chômage et taux court aux États-Unis, 1959–2009 : régimes et prévision hors échantillon\n\n## Conclusion\n\n**Ni la courbe de Phillips ni la règle de politique monétaire ne sont stables sur la période.**\n\n- **Phillips :** la rupture principale date de **1981T4**. Elle porte sur le niveau et la persistance de l'inflation, pas sur la pente du chômage prise seule. Elle est significative : sup-Wald de 44,4 contre un seuil bootstrap à 5 % de 33,5, p = 0,004. La pente seule ne rompt pas (p = 0,43).\n- **Règle de Taylor :** les ruptures sont plus incertaines. Le test global est marginal (p = 0,076). Bai-Perron suggère 1980T4 et 1985T1.\n- **Inflation :** ses moyennes basculent en **1973T1** et **1981T4** (PELT, robuste aux deux pénalités testées). Un régime volatil absorbe 1973–1983 puis réapparaît en 2008–09.\n\n**Pour la prévision, la variable à suivre est la tendance d'inflation, pas le chômage.** Sur 1980T1–2009T3, aucun modèle ne bat significativement l'AR de référence. Les modèles « économiques » (Phillips, VAR, Ridge) font moins bien que lui. Le meilleur prévisionniste ponctuel est le **modèle à tendance et volatilité stochastique (UCSV)** :\n\n| RMSFE relatif à l'AR | h = 1 | h = 4 | h = 8 |\n|---|---|---|---|\n| UCSV | 0,950 | 0,926 | 1,044 |\n| p de Diebold-Mariano | 0,18 | 0,25 | 0,73 |\n\nAprès 1985, l'avantage de UCSV sur l'AR est plus net : RMSFE relatif de 0,93, 0,84 et 0,82 aux trois horizons. Il reste à la limite de la significativité (p = 0,13, 0,07 et 0,11).\n\n## 1. Données et échantillon\n\nLe fichier compte 203 trimestres (1959T1–2009T3) et 13 colonnes, sans valeur manquante. Deux contrôles ont changé le traitement des données :\n\n- **Inflation recalculée :** `infl` correspond à 400·Δlog(CPI), à l'arrondi près. La première observation vaut 0, c'est un artefact sans retard disponible. `realint` vaut taux − inflation, sauf à cette même ligne. Je supprime donc 1959T1. L'échantillon utile court de 1959T2 à 2009T3, soit n = 202.\n- **Données de dernière vintage :** elles ne sont pas en temps réel. Le chômage et l'IPC sont peu révisés, ce qui limite le biais. Les conclusions de prévision s'entendent donc « pseudo hors échantillon ».\n\nLa cible de prévision est l'inflation annualisée moyenne sur les h trimestres suivants, avec h = 1, 4 et 8. Les prévisions sont directes : à l'origine t, seules les paires (X_s, y_{s+h}) avec s + h ≤ t servent à l'estimation. Il y a 119, 116 et 112 origines selon l'horizon (de 1979T4 à 2009T2 pour h = 1).\n\n## 2. Stationnarité\n\nLe verdict dépend du test et de la sous-période (tableau complet dans `artifacts/stationarity.csv`).\n\n| Série | Échantillon | ADF (p) | PP (p) | DF-GLS (p) | KPSS (p) |\n|---|---|---|---|---|---|\n| Inflation | complet | 0,027 | 0,000 | 0,006 | 0,081 |\n| Inflation | 1959–1983 | 0,456 | 0,002 | 0,107 | 0,010 |\n| Inflation | 1984–2009 | 0,000 | 0,000 | 0,000 | 0,042 |\n| Chômage | complet | 0,111 | 0,257 | 0,029 | 0,100 |\n| Taux 3 mois | complet | 0,280 | 0,298 | 0,099 | 0,054 |\n\n- **Inflation :** ADF, PP et DF-GLS rejettent la racine unitaire sur l'échantillon complet. Avant 1984, l'ADF ne la rejette pas (p = 0,456) et KPSS rejette la stationnarité (p = 0,010). Le test de Zivot-Andrews ne rejette pas non plus (p = 0,233, rupture estimée en 1972T2).\n- **Chômage et taux court :** ils se comportent comme des séries à racine presque unitaire. Les différences premières sont stationnaires (ADF p < 0,001).\n- **Test de Phillips-Perron :** il sur-rejette ici, à cause de la forte composante MA(1) négative de l'inflation trimestrielle.\n- **Lecture :** l'inflation ressemble à un processus stationnaire autour d'une moyenne qui change, pas à une racine unitaire pure. Cela justifie de tester les ruptures et de comparer des modèles à tendance variable.\n\n## 3. Ruptures structurelles (bibliothèque `ruptures` 1.1.10)\n\nTrois approches ont été croisées :\n\n- **Moyennes univariées :** PELT, Dynp avec sélection par BIC, Binseg et BottomUp, avec un segment minimal de 16 trimestres.\n- **Régressions :** Dynp avec coût linéaire (Bai-Perron) sur la courbe de Phillips ADL(2) et sur une règle de Taylor à ajustement partiel.\n- **Tests de rupture :** sup-Wald avec bootstrap sauvage récursif (élagage de 15 %, 499 tirages pour les p-values, 399 pour les seuils tracés). S'y ajoutent Chow et OLS-CUSUM.\n\nLes deux équations estimées sont les suivantes :\n- Phillips : π_t sur π_{t−1}, π_{t−2} et u_{t−1}.\n- Taylor : i_t sur i_{t−1}, l'inflation moyenne sur 4 trimestres et u_t.\n\nLe graphique ci-dessous montre les séries et les ruptures de moyenne PELT.\n\n![Séries et ruptures de moyenne PELT : l'inflation bascule en 1973T1 et 1981T4](/figures/fig-1795213300346-hevonmefj-0.png)\n\n**Inflation :** 1973T1 et 1981T4 ressortent avec les pénalités 2·ln n et 4·ln n, ainsi qu'avec Binseg. Dynp/BIC retient cinq ruptures (1965T4, 1973T1, 1977T4, 1981T4, 1991T1), mais les deux premières dates sont les plus stables.\n\n**Chômage et taux court :** PELT trouve plus de paliers, par exemple 1966T1, 1978T3, 1982T3, 1991T1 et 2001T3 pour le taux. Le BIC en retient 6 et 5 par construction sur des séries persistantes. Je les traite comme descriptives. Un BIC multivarié n'a jamais atteint de minimum pour k ≤ 6, et je l'ai abandonné.\n\nLa statistique de Wald en fonction de la date de rupture donne les résultats suivants.\n\n![Statistique sup-Wald et seuils bootstrap à 5 % pour la courbe de Phillips et la règle de Taylor](/figures/fig-1795213300745-39k6qboz5-0.png)\n\n| Test (bootstrap sauvage) | sup-Wald | Date | p |\n|---|---|---|---|\n| Phillips, tous coefficients | 44,4 | 1981T4 | 0,004 |\n| Phillips, intercept + persistance | 43,4 | 1981T4 | 0,002 |\n| Phillips, pente du chômage seule | 4,9 | 1982T3 | 0,43 |\n| Taylor, tous coefficients | 47,9 | 1980T3 | 0,076 |\n| Taylor, pente seule (inflation et chômage) | 10,5 | 1980T3 | 0,77 |\n\n- **Phillips :** la courbe est instable, mais pas parce que la pente du chômage s'effondre toute seule. C'est le niveau et la persistance de l'inflation qui changent en 1981T4.\n- **Taylor :** le test reste sous le seuil à 5 % (47,9 contre 58,9). OLS-CUSUM ne rejette ni Phillips (p = 0,26) ni Taylor (p = 0,39), faute de puissance.\n- **Chow :** pour Phillips, il rejette en 1984T1 (F = 6,45, p = 0,0001) et en 1979T3 (p = 0,019). Ces p-values sont asymptotiques et non robustes à l'hétéroscédasticité.\n\nLe graphique suivant donne les coefficients par régime, avec des erreurs HAC.\n\n![Pente de Phillips et réponse de long terme de la règle de Taylor par régime, IC 95 %](/figures/fig-1795213322098-9ha6rfkku-0.png)\n\n| Régime Phillips | n | Pente u(t−1) | Persistance (Σφ) |\n|---|---|---|---|\n| 1959T4–1972T4 | 53 | −0,55 (0,17) | 0,62 |\n| 1973T1–1981T3 | 35 | −1,08 (0,21) | 0,42 |\n| 1981T4–2009T3 | 112 | +0,10 (0,10) | 0,12 |\n\n**Phillips :** sur l'échantillon complet la pente vaut 0,01 (erreur-type 0,10). C'est la moyenne de régimes où elle est nettement négative (−0,55 puis −1,08) et d'un régime où elle est nulle. Après 1981T4, l'inflation trimestrielle devient presque sans mémoire (Σφ = 0,12). Aucune relation de Phillips n'est détectable à cette fréquence.\n\n**Taylor :** la réponse de long terme à l'inflation est de 0,67 (erreur-type 0,06) sur 1960T1–1980T3. Elle est de 0,81 sur 1980T4–1984T4, une fenêtre de seulement 17 trimestres. Elle est de 1,25 sur 1985T1–2009T3, mais avec une erreur-type de 1,05, car le lissage ρ = 0,96 frôle la racine unitaire. Le principe de Taylor (coefficient supérieur à 1) n'est donc pas démontrable après 1985. Avant 1980, il est rejeté : le coefficient reste nettement inférieur à 1. Le coefficient post-1985 est peu informatif.\n\nLes estimations sur fenêtres de 40 trimestres confirment l'instabilité.\n\n![Coefficients de Phillips et de Taylor sur fenêtres glissantes de 40 trimestres](/figures/fig-1795213314712-gkdlkqd20-0.png)\n\n- **Phillips :** la pente oscille entre −1,56 (1984T2) et +0,16 (2005T3). Elle n'est significative à 5 % que dans 27 % des fenêtres.\n- **Taylor :** le coefficient de long terme explose dès que ρ → 1, de −32 à +14 dans les cas extrêmes. Il est tronqué sur le graphique. Cette instabilité reflète d'abord un problème d'identification, et seulement ensuite un changement de politique.\n\nDeux limites s'ajoutent. Les régressions de la règle de Taylor sont estimées par MCO sans instruments, donc avec une endogénéité possible. Le chômage sert de proxy à l'écart de production.\n\n## 4. Régimes : Markov-switching et tendance d'inflation\n\nUn MS-AR(1) à deux régimes, avec intercept et variance commutants, est estimé sur l'échantillon complet.\n\n- **Régime volatil :** moyenne conditionnelle de 5,98 %, écart-type de 3,69 points, durée attendue de 16 trimestres.\n- **Régime calme :** moyenne de 2,92 %, écart-type de 1,27 point, durée attendue de 30 trimestres.\n- **Probabilité lissée du régime volatil :** 0,04 en 1959–72, 0,93 en 1973–83, 0,22 en 1984–2007 et 0,92 en 2008–09.\n\nLa détection de ce modèle recoupe celle de `ruptures`, mais la « Grande Modération » apparaît comme un retour progressif à un régime calme, pas comme une rupture unique en 1984. La tendance UCSV (filtre à particules, γ = 0,2, estimée en temps réel) passe de 14,1 % en 1980T1 à 4,8 % en 1990T1, 3,1 % en 2000T1 et 3,3 % fin 2007.\n\n![Tendance d'inflation UCSV et probabilité lissée du régime volatil](/figures/fig-1795213337724-l98nh8kv9-0.png)\n\n## 5. Prévision hors échantillon\n\n**Modèles comparés :** treize modèles, plus une combinaison égale de cinq d'entre eux (AR, Phillips + taux, VAR, MS-AR(1), UCSV).\n- Références : marche aléatoire sur le dernier trimestre (RW1) et sur la moyenne de 4 trimestres (RW4), AR(p ≤ 4, BIC), IMA(1,1).\n- Courbes de Phillips : trois variantes.\n- Systèmes : VAR à 3 variables (inflation, chômage, taux), avec p choisi par BIC.\n- Régimes : Markov-switching AR(1) et AR(3) ; le MS-PC (Phillips à régimes) n'est évalué qu'à h = 1.\n- Autres : tendance UCSV, Ridge sur 40 prédicteurs.\n\n**Évaluation :** fenêtre extensible comme référence. Fenêtres glissantes de 60 et 40 trimestres en robustesse. Tests de Diebold-Mariano avec correction de Harvey-Leybourne-Newbold, perte quadratique, et Clark-West pour les modèles emboîtés. Les estimations lourdes ont tourné en parallèle (`code/fc_linear.py`, `fc_ms.py`, `fc_ms3.py`, `fc_ucsv.py`).\n\n| RMSFE (fenêtre extensible) | h = 1 | h = 4 | h = 8 |\n|---|---|---|---|\n| AR (BIC) | 2,595 | 1,823 | **1,629** |\n| UCSV | **2,466** | **1,688** | 1,701 |\n| IMA(1,1) | 2,567 | 1,821 | 1,740 |\n| RW4 | 2,591 | 1,842 | 1,779 |\n| Combinaison égale | 2,545 | 1,862 | 1,824 |\n| MS-AR(3) | 2,643 | 1,844 | 1,739 |\n| Phillips, inflation + chômage | 2,626 | 1,841 | 1,733 |\n| MS-AR(1) | 2,711 | 2,014 | 1,799 |\n| Phillips ADL, 2 retards de u | 2,730 | 1,874 | 1,883 |\n| VAR(3 variables) | 2,808 | 2,296 | 2,476 |\n| Ridge | 2,683 | 2,295 | 2,392 |\n| Phillips + taux court | 2,843 | 2,393 | 2,662 |\n| RW1 | 3,020 | 2,411 | 2,226 |\n\n![RMSFE relatif à l'AR, fenêtre extensible, 1980T1–2009T3](/figures/fig-1795213353082-wzprr6cat-0.png)\n\n**Résultats principaux :**\n\n- **UCSV :** meilleur à h = 1 et 4, mais non significatif face à l'AR (p DM = 0,18 et 0,25). À h = 8, l'AR garde le meilleur RMSFE absolu.\n- **Phillips, VAR, Ridge :** tous plus mauvais que l'AR. Les écarts les plus marqués (VAR et Phillips + taux à h = 4, p ≈ 0,08) vont dans le mauvais sens.\n- **Clark-West :** il ne trouve de contenu informatif ajouté qu'une fois, pour Ridge à h = 1 (p = 0,041). Les modèles de Phillips ne passent pas ce test.\n- **Granger dans le VAR(2) :** le chômage ne cause pas l'inflation (p = 0,976). Le taux court le fait (p = 0,006). Le chômage n'apporte aucune information prévisionnelle ; le taux court peut en apporter dans le système estimé, mais cela ne se traduit pas en gain hors échantillon.\n- **Variantes de Phillips :** Δu, écart de chômage à la moyenne mobile et taux réel donnent des RMSFE relatifs de 1,02, 1,015 et 1,05 à h = 1. Aucune ne bat l'AR.\n- **Markov-switching :** ne bat pas l'AR non plus, ni en AR(1) ni en AR(3).\n- **Model Confidence Set (10 %) :** sur l'échantillon complet, il garde les onze modèles à tous les horizons. Les données ne permettent pas de classer les modèles avec confiance.\n\n**Cohérence avec la littérature :** ce résultat est cohérent avec l'idée qu'à fréquence trimestrielle et sur un échantillon court, la prévision d'inflation est dominée par la persistance et par la tendance, pas par les écarts d'activité.\n\nLe gain de UCSV est concentré dans le temps.\n\n![Gain cumulé d'erreur quadratique par rapport à l'AR, h = 4](/figures/fig-1795213363664-5zlb9f2a6-0.png)\n\nLe gain cumulé de UCSV sur l'AR à h = 4 est de −20,7 fin 1989 : il perd 20,7 sur la période 1980–89. Il remonte à +22,0 fin 2007, puis à +54,9 après la crise (le dernier point, origine 2008T3, inclut la déflation de 2008T4). Le VAR perd 218 points à la même date de 1989. Les erreurs de 2008–09 pèsent lourd : 25,6 % de la somme des carrés des erreurs d'AR à h = 1 viennent des origines de 2008T3 et après.\n\n## 6. Robustesse\n\nLes résultats suivants valent pour l'AR, les modèles de Phillips, le VAR et UCSV. Ils sont détaillés dans `artifacts/`.\n\n![RMSFE relatif à l'AR par sous-période, h = 1, 4 et 8](/figures/fig-1795213375061-fw6xmpnym-0.png)\n\n| Variation du protocole | Ce qui change | Verdict |\n|---|---|---|\n| Origines 1980–1989 | AR difficile à battre ; PC + taux et VAR à 1,5–1,6 à h = 4 | Le résultat « les modèles économiques sont plus mauvais » est surtout dû aux années 1980 |\n| Origines 1990–2007 | UCSV 0,84 à h = 4 (p = 0,094), 0,75 à h = 8 (p = 0,022) ; IMA(1,1) et RW4 à ≈ 0,81 à h = 8 | Les modèles à moyenne ancrée gagnent après 1990 |\n| Origines à partir de 1985T1 | UCSV 0,93 / 0,84 / 0,82 ; MCS à h = 8 : Combinaison, IMA, MS-AR(3), RW4, Ridge, UCSV (AR, Phillips, VAR exclus) | L'AR n'est plus dans l'ensemble des meilleurs à h = 8 |\n| Hors crise (cibles avant 2008T3) | UCSV 1,015 / 0,979 / 1,067 | L'avantage de UCSV à h = 1 disparaît sans la crise |\n| Perte MAE | UCSV 0,967 / 0,950 / 0,938 | Même classement avec une perte moins sensible aux valeurs extrêmes |\n| Fenêtres glissantes (60 et 40) | IMA(1,1) et UCSV gagnent aussi à h = 8 (rapport AR de 0,98 à 0,87) ; l'AR se dégrade (1,735 à h = 8 en fenêtre de 60) | La fenêtre extensible favorise l'AR à h = 8 |\n| MS-AR(3) au lieu de MS-AR(1) | RMSFE 2,643 / 1,844 / 1,739, contre 2,711 / 2,014 / 1,799 | Mieux spécifié, mais toujours pas meilleur que l'AR |\n\nLe résultat qui tient le plus est négatif : Phillips, VAR et Ridge ne battent jamais l'AR, quel que soit le protocole. Le résultat positif (UCSV) dépend du protocole : fort après 1985, faible sur l'échantillon complet.\n\n**Limites de la comparaison :**\n- **Fenêtres glissantes :** les modèles Markov-switching n'y convergent pas toujours. Pour MS-AR(1), il y a 2 échecs avec 60 trimestres et 6 avec 40. Pour MS-PC, 16 et 28. Pour MS-AR(3) en fenêtre de 60, 13. Ces origines sont laissées vides et exclues des échantillons communs.\n- **Tests multiples :** je teste beaucoup de modèles, d'horizons et de périodes. Les p-values nominales ne sont pas corrigées. Le résultat UCSV à h = 8 sur 1990–2007 (p = 0,022) est le plus fort, mais il vient de la sélection d'une sous-période après coup.\n\n## 7. Diagnostics des résidus\n\n![Autocorrélation des résidus de l'AR(3) et de la courbe de Phillips ADL(2)](/figures/fig-1795213375401-hoof4mdm0-0.png)\n\n| Modèle | Ljung-Box(4) p | ARCH(4) p | Kurtosis |\n|---|---|---|---|\n| AR(3) | 0,951 | < 0,001 | 7,4 |\n| Phillips ADL(2) | 0,004 | < 0,001 | 6,7 |\n| Règle de Taylor | 0,006 | < 0,001 | 17,6 |\n| VAR(2), équation inflation | 0,006 | < 0,001 | 6,6 |\n| MS-AR(1), résidus standardisés | < 0,001 | 0,688 | 6,5 |\n\n- **AR(3) :** il passe le test d'autocorrélation, mais pas l'hétéroscédasticité. Le VAR(2) est stable (plus grande racine : module 0,955), mais son test de blancheur est rejeté.\n- **Phillips ADL(2) :** il laisse de l'autocorrélation, ce qui montre une dynamique insuffisante.\n- **Règle de Taylor :** son excès de kurtosis (17,6) vient de la période Volcker.\n- **MS-AR(1) :** il supprime l'effet ARCH (p = 0,688) mais laisse de l'autocorrélation, car un seul retard est trop court. C'est ce qui a motivé le MS-AR(3).\n- **Conséquence :** les écarts-types classiques sont trop optimistes. Je me suis appuyé sur HAC, le bootstrap sauvage et des tests de prévision robustes à l'autocorrélation.\n- **Biais hors échantillon à h = 1 :** il est de −0,34 point pour l'AR (p = 0,049), qui sur-prédit en moyenne, et de −0,25 pour UCSV (p = 0,18).\n\n## 8. Limites et ce qui a échoué\n\n- **Données en dernière vintage :** les conclusions sont « pseudo hors échantillon », pas en temps réel.\n- **Échantillon court :** 119 origines à h = 1, avec des erreurs fortement chevauchantes à h = 4 et 8. La puissance est faible : aucun test d'égalité de précision n'atteint 5 % en échantillon complet.\n- **Crise 2008–09 :** elle domine les erreurs et fait varier le résultat.\n- **Règle de Taylor :** elle est estimée avec un proxy du chômage, sans écart de production en temps réel, et sans variables instrumentales.\n- **Ce qui n'a pas marché :**\n  - Une première installation de `ruptures` par pip a échoué (certificat SSL) ; `install_package` a fonctionné.\n  - Le BIC multivarié de `ruptures` n'a jamais atteint de minimum (abandonné).\n  - Une première version du script Markov-switching lisait les intercepts dans de mauvaises positions de paramètres ; j'ai corrigé avant toute exécution.\n  - Les modèles à régimes divergent souvent en fenêtre glissante (échecs listés plus haut).\n- **Ce que je n'ai pas fait :** un test de Giacomini-Rossi sur la stabilité de la précision relative (je ne vérifie pas ses valeurs critiques), et un UCSV avec estimation bayésienne complète des hyperparamètres. Je les ai fixés à γ = 0,2 et 0,1 ; les résultats sont proches (RMSFE de 2,495 à h = 1 avec γ = 0,1).\n\n## Fichiers\n\n- **Code :** `code/common.py`, `code/fcsuite.py`, `code/fc_linear.py`, `code/fc_ms.py`, `code/fc_ms3.py`, `code/fc_ucsv.py`, `code/evalfc.py` (aussi dans la bibliothèque : `from vqlib.fc_eval import dm_test, cw_test`).\n- **Prévisions brutes :** `artifacts/forecasts/`.\n- **Tableaux :**\n  - `artifacts/stationarity.csv`\n  - `artifacts/pc_regimes.csv`\n  - `artifacts/taylor_regimes.csv`\n  - `artifacts/rmsfe_expanding.csv`\n  - `artifacts/dm_vs_AR_expanding.csv`\n  - `artifacts/subsample_results.csv`\n  - `artifacts/mcs_results.csv`\n  - `artifacts/robust_window_rmsfe.csv`\n  - `artifacts/residual_diagnostics.csv`\n- **Figures vectorielles :** `artifacts/figures/`.\n- Le registre des expériences, échecs compris, est dans le workspace.\n\n**Sources :** fichier téléversé `us_macro_quarterly.csv` ; calculs de l'auteur avec statsmodels, arch et ruptures.\n\n## Correctifs\n\n**Mot « significative » (sections 3 et 5).** Les p-values associées sont dans le registre (expériences sup-Wald, Diebold-Mariano et MCS). Une phrase n'en avait pas : « la pente n'est significative à 5 % que dans 27 % des fenêtres ». Ce 27 % est la part des fenêtres glissantes de 40 trimestres où |t| > 1,96 (erreurs HAC). Ce n'est pas une p-value et ce n'est pas un test formel : les fenêtres se chevauchent fortement, donc elles ne sont pas indépendantes. Cette part décrit une instabilité, elle ne la teste pas. Je l'ai ajoutée au registre (expérience n° 65) avec cette réserve, ainsi que les extrêmes de pente (−1,56 en 1984T2, +0,16 en 2005T3).\n\n**Défauts de figures.** C'est une fausse alerte. Les trois figures signalées sont les premières versions, dont les textes se chevauchaient. Je les ai redessinées avant la rédaction et le rapport n'utilise que les versions corrigées :\n- la figure des coefficients par régime ;\n- la figure sur la tendance UCSV et les régimes markoviens ;\n- la figure des gains cumulés d'erreur quadratique.\n\nLe contrôle automatique n'a signalé aucun défaut sur ces versions corrigées. Les versions défectueuses restent seulement dans `artifacts/figures/` ; le rapport ne renvoie pas vers elles.","toolResults":[{"tool":"inspect_dataset","result":{"path":"data/us_macro_quarterly.csv","format":"csv","bytes":17424,"rows":203,"rowsProfiled":203,"truncated":false,"columns":[{"name":"period","dtype":"str","missing":0,"missingShare":0,"distinct":203,"examples":["1959Q1","1959Q2","1959Q3"],"parsesAsDate":true,"range":["1959-01-01 00:00:00","2009-07-01 00:00:00"]},{"name":"realgdp","dtype":"float64","missing":0,"missingShare":0,"distinct":203,"examples":[2710.349,2778.801,2775.488],"summary":{"mean":7221.171901477834,"std":3214.956043957166,"min":2710.349,"25%":4440.1035,"50%":6559.594,"75%":9629.3465,"max":13415.266},"zeros":0},{"name":"realcons","dtype":"float64","missing":0,"missingShare":0,"distinct":202,"examples":[1707.4,1733.7,1751.8],"summary":{"mean":4825.293103448276,"std":2313.346192143467,"min":1707.4,"25%":2874.1000000000004,"50%":4299.9,"75%":6398.15,"max":9363.6},"zeros":0},{"name":"realinv","dtype":"float64","missing":0,"missingShare":0,"distinct":203,"examples":[286.898,310.859,289.226],"summary":{"mean":1012.8638620689655,"std":585.1022673538571,"min":259.764,"25%":519.1475,"50%":896.21,"75%":1436.6815000000001,"max":2264.721},"zeros":0},{"name":"realgovt","dtype":"float64","missing":0,"missingShare":0,"distinct":203,"examples":[470.045,481.301,491.26],"summary":{"mean":663.3286403940887,"std":140.8636547185507,"min":460.4,"25%":527.9594999999999,"50%":662.412,"75%":773.049,"max":1044.088},"zeros":0},{"name":"realdpi","dtype":"float64","missing":0,"missingShare":0,"distinct":203,"examples":[1886.9,1919.7,1916.4],"summary":{"mean":5310.540886699508,"std":2423.51597676942,"min":1886.9,"25%":3276.95,"50%":4959.4,"75%":6977.85,"max":10077.5},"zeros":0},{"name":"cpi","dtype":"float64","missing":0,"missingShare":0,"distinct":203,"examples":[28.98,29.15,29.35],"summary":{"mean":105.07578817733989,"std":61.27887772937541,"min":28.98,"25%":41.05,"50%":104.1,"75%":159.65,"max":218.61},"zeros":0},{"name":"m1","dtype":"float64","missing":0,"missingShare":0,"distinct":200,"examples":[139.7,141.7,140.5],"summary":{"mean":667.9275862068966,"std":455.3463814032303,"min":139.6,"25%":228.64999999999998,"50%":540.9,"75%":1102.1,"max":1673.9},"zeros":0},{"name":"tbilrate","dtype":"float64","missing":0,"missingShare":0,"distinct":183,"examples":[2.82,3.08,3.82],"summary":{"mean":5.311773399014779,"std":2.803070771560717,"min":0.12,"25%":3.5149999999999997,"50%":5.01,"75%":6.665,"max":15.33},"zeros":0},{"name":"unemp","dtype":"float64","missing":0,"missingShare":0,"distinct":56,"examples":[5.8,5.1,5.3],"summary":{"mean":5.8847290640394085,"std":1.4585742735215175,"min":3.4,"25%":4.9,"50%":5.7,"75%":6.8,"max":10.7},"zeros":0},{"name":"pop","dtype":"float64","missing":0,"missingShare":0,"distinct":203,"examples":[177.146,177.83,178.657],"summary":{"mean":239.7241527093596,"std":37.39044963546054,"min":177.146,"25%":208.631,"50%":236.348,"75%":271.7215,"max":308.013},"zeros":0},{"name":"infl","dtype":"float64","missing":0,"missingShare":0,"distinct":179,"examples":[0,2.34,2.74],"summary":{"mean":3.9613300492610835,"std":3.2532164897439695,"min":-8.79,"25%":2.2699999999999996,"50%":3.24,"75%":4.975,"max":14.62},"zeros":1},{"name":"realint","dtype":"float64","missing":0,"missingShare":0,"distinct":177,"examples":[0,0.74,1.09],"summary":{"mean":1.3365024630541873,"std":2.668798647618167,"min":-6.79,"25%":-0.08499999999999999,"50%":1.34,"75%":2.63,"max":10.95},"zeros":1}],"columnCount":13,"duplicatedRows":0,"timeColumns":["period"],"index":"int64","memoryMb":0.03,"head":[{"period":"1959Q1","realgdp":2710.349,"realcons":1707.4,"realinv":286.898,"realgovt":470.045,"realdpi":1886.9,"cpi":28.98,"m1":139.7,"tbilrate":2.82,"unemp":5.8,"pop":177.146,"infl":0,"realint":0},{"period":"1959Q2","realgdp":2778.801,"realcons":1733.7,"realinv":310.859,"realgovt":481.301,"realdpi":1919.7,"cpi":29.15,"m1":141.7,"tbilrate":3.08,"unemp":5.1,"pop":177.83,"infl":2.34,"realint":0.74},{"period":"1959Q3","realgdp":2775.488,"realcons":1751.8,"realinv":289.226,"realgovt":491.26,"realdpi":1916.4,"cpi":29.35,"m1":140.5,"tbilrate":3.82,"unemp":5.3,"pop":178.657,"infl":2.74,"realint":1.09},{"period":"1959Q4","realgdp":2785.204,"realcons":1753.7,"realinv":299.356,"realgovt":484.052,"realdpi":1931.3,"cpi":29.37,"m1":140,"tbilrate":4.33,"unemp":5.6,"pop":179.386,"infl":0.27,"realint":4.06},{"period":"1960Q1","realgdp":2847.699,"realcons":1770.5,"realinv":331.722,"realgovt":462.199,"realdpi":1955.5,"cpi":29.54,"m1":139.6,"tbilrate":3.5,"unemp":5.2,"pop":180.007,"infl":2.31,"realint":1.19},{"period":"1960Q2","realgdp":2834.39,"realcons":1792.9,"realinv":298.152,"realgovt":460.4,"realdpi":1966.1,"cpi":29.55,"m1":140.2,"tbilrate":2.68,"unemp":5.2,"pop":180.671,"infl":0.14,"realint":2.55},{"period":"1960Q3","realgdp":2839.022,"realcons":1785.8,"realinv":296.375,"realgovt":474.676,"realdpi":1967.8,"cpi":29.75,"m1":140.9,"tbilrate":2.36,"unemp":5.6,"pop":181.528,"infl":2.7,"realint":-0.34},{"period":"1960Q4","realgdp":2802.616,"realcons":1788.2,"realinv":259.764,"realgovt":476.434,"realdpi":1966.6,"cpi":29.84,"m1":141.1,"tbilrate":2.29,"unemp":6.3,"pop":182.287,"infl":1.21,"realint":1.08}]}},{"tool":"research_memory","result":{"notes":[],"kinds":["dataset","model","decision","implementation","failure","hypothesis","finding","workflow"]}},{"tool":"update_plan","result":"Plan shown to the user."},{"tool":"execute_custom_python_analysis","result":{"success":true,"output":"(203, 13) 1959-03-31 00:00:00 2009-09-30 00:00:00\ncount    202.000000\nmean       0.000155\nstd        0.002762\nmin       -0.004973\n25%       -0.001837\n50%        0.000099\n75%        0.002468\nmax        0.004876\ndtype: float64\n2.82\n          count      mean       std   min    25%   50%    75%    max\ninfl      203.0  3.961330  3.253216 -8.79  2.270  3.24  4.975  14.62\nunemp     203.0  5.884729  1.458574  3.40  4.900  5.70  6.800  10.70\ntbilrate  203.0  5.311773  2.803071  0.12  3.515  5.01  6.665  15.33\nrealint   203.0  1.336502  2.668799 -6.79 -0.085  1.34  2.630  10.95\ndate\n92.0    101\n91.0     64\n90.0     37\nName: count, dtype: int64\n0\n            infl  unemp  tbilrate\ndate                             \n2008-03-31  2.82    4.9      1.56\n2008-06-30  8.53    5.4      1.74\n2008-09-30 -3.16    6.0      1.17\n2008-12-31 -8.79    6.9      0.12\n2009-03-31  0.94    8.1      0.22\n2009-06-30  3.37    9.2      0.18\n2009-09-30  3.56    9.6      0.12\n\n[Notebook session — variables in memory: df: DataFrame 203x13, calc: Series 203]","result":null,"figures":[],"files":[]}},{"tool":"execute_custom_python_analysis","result":{"success":true,"output":"Installed ruptures. Import it now. Environment recorded in environment.lock.txt.\n1.1.10\n\n[Notebook session — variables in memory: df: DataFrame 203x13, calc: Series 203]","result":null,"figures":[],"files":[]}},{"tool":"execute_custom_python_analysis","result":{"success":true,"output":"var     sample    n  ADF_p   PP_p  DFGLS_p  KPSS_stat  KPSS_p\n0       infl       full  202  0.027  0.000    0.006      0.392   0.081\n1       infl  1959-1983   99  0.456  0.002    0.107      1.028   0.010\n2       infl  1984-2009  103  0.000  0.000    0.000      0.498   0.042\n3      unemp       full  202  0.111  0.257    0.029      0.233   0.100\n4      unemp  1959-1983   99  0.118  0.660    0.017      0.903   0.010\n5      unemp  1984-2009  103  0.361  0.329    0.124      0.483   0.046\n6   tbilrate       full  202  0.280  0.298    0.099      0.454   0.054\n7   tbilrate  1959-1983   99  0.603  0.396    0.447      1.223   0.010\n8   tbilrate  1984-2009  103  0.573  0.392    0.453      1.181   0.010\n9    realint       full  202  0.008  0.000    0.001      0.160   0.100\n10   realint  1959-1983   99  0.436  0.000    0.087      0.220   0.100\n11   realint  1984-2009  103  0.079  0.000    0.176      1.031   0.010\n12    growth       full  201  0.000  0.000    0.000      0.278   0.100\n13    growth  1959-1983   98  0.000  0.000    0.000      0.181   0.100\n14    growth  1984-2009  103  0.000  0.000    0.056      0.428   0.065\ninfl ZA stat -4.21 p 0.233 break idx 1972Q2\nunemp ZA stat -3.73 p 0.526 break idx 1973Q4\ntbilrate ZA stat -3.63 p 0.597 break idx 1977Q1\ninfl diff ADF p 0.0\nunemp diff ADF p 0.0008\ntbilrate diff ADF p 0.0\nStderr: actual p-value is greater than the p-value returned.\n<analysis>:24: InterpolationWarning: The test statistic is outside of the range of p-values available in the\nlook-up table. The actual p-value is smaller than the p-value returned.\n<analysis>:24: InterpolationWarning: The test statistic is outside of the range of p-values available in the\nlook-up table. The actual p-value is smaller than the p-value returned.\n<analysis>:24: InterpolationWarning: The test statistic is outside of the range of p-values available in the\nlook-up table. The actual p-value is smaller than the p-value returned.\n<analysis>:24: InterpolationWarning: The test statistic is outside of the range of p-values available in the\nlook-up table. The actual p-value is greater than the p-value returned.\n<analysis>:24: InterpolationWarning: The test statistic is outside of the range of p-values available in the\nlook-up table. The actual p-value is greater than the p-value returned.\n<analysis>:24: InterpolationWarning: The test statistic is outside of the range of p-values available in the\nlook-up table. The actual p-value is smaller than the p-value returned.\n<analysis>:24: InterpolationWarning: The test statistic is outside of the range of p-values available in the\nlook-up table. The actual p-value is greater than the p-value returned.\n<analysis>:24: InterpolationWarning: The test statistic is outside of the range of p-values available in the\nlook-up table. The actual p-value is greater than the p-value returned.\n\n[Notebook session — variables in memory: df: DataFrame 203x13, calc: Series 203, load(), d: DataFrame 202x15, kpss(), zivot_andrews(), DFGLS(), PhillipsPerron(), rows: list[15], v: str, x: Series 201, name: str, sub: Series 103, adf: tuple[6], k: tuple[4], dg: float64, pp: float64, st: DataFrame 15x8, za: tuple[5]]","result":null,"figures":[],"files":[]}},{"tool":"execute_custom_python_analysis","result":{"success":true,"output":"1267\n\n[Notebook session — variables in memory: df: DataFrame 203x13, calc: Series 203, load(), d: DataFrame 202x15, kpss(), zivot_andrews(), DFGLS(), PhillipsPerron(), rows: list[15], v: str, x: Series 201, name: str, sub: Series 103, adf: tuple[6], k: tuple[4], dg: float64, pp: float64, st: DataFrame 15x8, za: tuple[5], code: str]","result":null,"figures":[],"files":[]}},{"tool":"execute_custom_python_analysis","result":{"success":true,"output":"3973\n\n[Notebook session — variables in memory: df: DataFrame 203x13, calc: Series 203, load(), d: DataFrame 202x15, kpss(), zivot_andrews(), DFGLS(), PhillipsPerron(), rows: list[15], v: str, x: Series 201, name: str, sub: Series 103, adf: tuple[6], k: tuple[4], dg: float64, pp: float64, st: DataFrame 15x8, za: tuple[5], code: str, s: str]","result":null,"figures":[],"files":[]}},{"tool":"research_run","result":{"run_id":"fe07fc39-611b-47f2-b7ca-f19c05d3dc73","label":"Linear models expanding","code":"code/fc_linear.py","status":"running","exit_code":null,"seconds":0,"progress":null,"error":null,"logs":"runs/fe07fc39-611b-47f2-b7ca-f19c05d3dc73/stdout.log, runs/fe07fc39-611b-47f2-b7ca-f19c05d3dc73/stderr.log","note":"Started in the background. Use research_runs (wait / status / logs) to follow it; start other runs meanwhile if they are independent."}},{"tool":"research_run","result":{"run_id":"908d8db9-e156-49bd-b54f-a3b0eee8db27","label":"Linear models rolling 60","code":"code/fc_linear.py","status":"running","exit_code":null,"seconds":0,"progress":null,"error":null,"logs":"runs/908d8db9-e156-49bd-b54f-a3b0eee8db27/stdout.log, runs/908d8db9-e156-49bd-b54f-a3b0eee8db27/stderr.log","note":"Started in the background. Use research_runs (wait / status / logs) to follow it; start other runs meanwhile if they are independent."}},{"tool":"research_run","result":{"run_id":"64827983-99cb-4ddf-a37e-2f7184f388d4","label":"Linear models rolling 40 (robustness)","code":"code/fc_linear.py","status":"running","exit_code":null,"seconds":0,"progress":null,"error":null,"logs":"runs/64827983-99cb-4ddf-a37e-2f7184f388d4/stdout.log, runs/64827983-99cb-4ddf-a37e-2f7184f388d4/stderr.log","note":"Started in the background. Use research_runs (wait / status / logs) to follow it; start other runs meanwhile if they are independent."}},{"tool":"execute_custom_python_analysis","result":{"success":true,"output":"1323\n\n[Notebook session — variables in memory: df: DataFrame 203x13, calc: Series 203, load(), d: DataFrame 202x15, kpss(), zivot_andrews(), DFGLS(), PhillipsPerron(), rows: list[15], v: str, x: Series 201, name: str, sub: Series 103, adf: tuple[6], k: tuple[4], dg: float64, pp: float64, st: DataFrame 15x8, za: tuple[5], code: str, s: str, ms: str, uc: str]","result":null,"figures":[],"files":[]}},{"tool":"execute_custom_python_analysis","result":{"success":true,"output":"['p[0->0]', 'p[1->0]', 'const[0]', 'const[1]', 'x1[1]', 'sigma2[0]', 'sigma2[1]']\n0.13400769233703613\n[ 0.966  0.062  1.376  2.813  0.529  1.607 13.639]\n[[0.96628658 0.0620335 ]\n [0.03371342 0.9379665 ]] [0.32366618 0.67633382]\n\n[Notebook session — variables in memory: df: DataFrame 203x13, calc: Series 203, load(), d: DataFrame 202x15, kpss(), zivot_andrews(), DFGLS(), PhillipsPerron(), rows: list[15], v: str, x: Series 201, name: str, sub: Series 103, adf: tuple[6], k: tuple[4], dg: float64, pp: float64, st: DataFrame 15x8, za: tuple[5], code: str, s: str, ms: str, uc: str, MarkovRegression(), y: ndarray (201,), X: ndarray (201, 1), mod: MarkovRegression, t: float, r: MarkovRegressionResultsWrapper]","result":null,"figures":[],"files":[]}},{"tool":"execute_custom_python_analysis","result":{"success":true,"output":"True\n\n[Notebook session — variables in memory: df: DataFrame 203x13, calc: Series 203, load(), d: DataFrame 202x15, kpss(), zivot_andrews(), DFGLS(), PhillipsPerron(), rows: list[15], v: str, x: Series 201, name: str, sub: Series 103, adf: tuple[6], k: tuple[4], dg: float64, pp: float64, st: DataFrame 15x8, za: tuple[5], code: str, s: str, ms: str, uc: str, MarkovRegression(), y: ndarray (201,), X: ndarray (201, 1), mod: MarkovRegression, t: float, r: MarkovRegressionResultsWrapper]","result":null,"figures":[],"files":[]}},{"tool":"research_run","result":{"run_id":"8125167d-cec8-434f-b799-c5365fcf2a6c","label":"Markov-switching expanding","code":"code/fc_ms.py","status":"running","exit_code":null,"seconds":0,"progress":null,"error":null,"logs":"runs/8125167d-cec8-434f-b799-c5365fcf2a6c/stdout.log, runs/8125167d-cec8-434f-b799-c5365fcf2a6c/stderr.log","note":"Started in the background. Use research_runs (wait / status / logs) to follow it; start other runs meanwhile if they are independent."}},{"tool":"research_run","result":{"run_id":"a0e1b107-631d-489d-a128-c8f32694c432","label":"Markov-switching rolling 60","code":"code/fc_ms.py","status":"running","exit_code":null,"seconds":0,"progress":null,"error":null,"logs":"runs/a0e1b107-631d-489d-a128-c8f32694c432/stdout.log, runs/a0e1b107-631d-489d-a128-c8f32694c432/stderr.log","note":"Started in the background. Use research_runs (wait / status / logs) to follow it; start other runs meanwhile if they are independent."}},{"tool":"research_run","result":{"run_id":"e887c62f-bb1d-45af-9fcd-47088bd65bbf","label":"UCSV particle filter gamma 0.2","code":"code/fc_ucsv.py","status":"running","exit_code":null,"seconds":0,"progress":null,"error":null,"logs":"runs/e887c62f-bb1d-45af-9fcd-47088bd65bbf/stdout.log, runs/e887c62f-bb1d-45af-9fcd-47088bd65bbf/stderr.log","note":"Started in the background. Use research_runs (wait / status / logs) to follow it; start other runs meanwhile if they are independent."}},{"tool":"research_run","result":{"run_id":"a2531baa-439f-4776-a522-baced94a1bf1","label":"UCSV particle filter gamma 0.1","code":"code/fc_ucsv.py","status":"running","exit_code":null,"seconds":0,"progress":null,"error":null,"logs":"runs/a2531baa-439f-4776-a522-baced94a1bf1/stdout.log, runs/a2531baa-439f-4776-a522-baced94a1bf1/stderr.log","note":"Started in the background. Use research_runs (wait / status / logs) to follow it; start other runs meanwhile if they are independent."}},{"tool":"research_runs","result":{"runs":[{"run_id":"fe07fc39-611b-47f2-b7ca-f19c05d3dc73","label":"Linear models expanding","code":"code/fc_linear.py","status":"done","exit_code":0,"seconds":20,"progress":1,"error":null,"logs":"runs/fe07fc39-611b-47f2-b7ca-f19c05d3dc73/stdout.log, runs/fe07fc39-611b-47f2-b7ca-f19c05d3dc73/stderr.log"},{"run_id":"908d8db9-e156-49bd-b54f-a3b0eee8db27","label":"Linear models rolling 60","code":"code/fc_linear.py","status":"done","exit_code":0,"seconds":20,"progress":1,"error":null,"logs":"runs/908d8db9-e156-49bd-b54f-a3b0eee8db27/stdout.log, runs/908d8db9-e156-49bd-b54f-a3b0eee8db27/stderr.log"},{"run_id":"64827983-99cb-4ddf-a37e-2f7184f388d4","label":"Linear models rolling 40 (robustness)","code":"code/fc_linear.py","status":"done","exit_code":0,"seconds":20,"progress":1,"error":null,"logs":"runs/64827983-99cb-4ddf-a37e-2f7184f388d4/stdout.log, runs/64827983-99cb-4ddf-a37e-2f7184f388d4/stderr.log"}]}},{"tool":"execute_custom_python_analysis","result":{"success":true,"output":"def fit(self, signal) -> \"CostLinear\":\n        \"\"\"Set parameters of the instance. The first column contains the\n        observed variable. The other columns contains the covariates.\n\n        Args:\n            signal (array): signal of shape (n_samples, n_regressors+1)\n\n        Returns:\n            self\n        \"\"\"\n        assert signal.ndim > 1, \"Not enough dimensions\"\n\n        self.signal = signal[:, 0].reshape(-1, 1)\n        self.covar = signal[:, 1:]\n        return self\n\n    def error(self, start, end) -> float:\n        \"\"\"Return the approximation cost on the segment [start:end].\n\n        Args:\n            start (int): start of the segment\n            end (int): end of the segment\n\n        Returns:\n            segment cost\n\n        Raises:\n            NotEnoughPoints: when the segment is too short (less than `min_size` samples).\n        \"\"\"\n        if end - start < self.min_size:\n            raise NotEnoughPoints\n        y, X = self.signal[start:end], self.covar[start:end]\n        _, residual, _, _ = lstsq(X, y, rcond=None)\n        return residual.sum()\n\n[Notebook session — variables in memory: df: DataFrame 203x13, calc: Series 203, load(), d: DataFrame 202x15, kpss(), zivot_andrews(), DFGLS(), PhillipsPerron(), rows: list[15], v: str, x: Series 201, name: str, sub: Series 103, adf: tuple[6], k: tuple[4], dg: float64, pp: float64, st: DataFrame 15x8, za: tuple[5], code: str, s: str, ms: str, uc: str, MarkovRegression(), y: ndarray (201,), X: ndarray (201, 1), mod: MarkovRegression, t: float, r: MarkovRegressionResultsWrapper, CostLinear()]","result":null,"figures":[],"files":[]}},{"tool":"execute_custom_python_analysis","result":{"success":true,"output":"infl BIC best k = 5 ['1965Q4', '1973Q1', '1977Q4', '1981Q4', '1991Q1'] | BIC by k [np.float64(480.4), np.float64(464.2), np.float64(364.8), np.float64(354.6), np.float64(351.7), np.float64(350.7), np.float64(357.0)]\nunemp BIC best k = 6 ['1965Q2', '1970Q2', '1974Q4', '1981Q4', '1985Q4', '1994Q4'] | BIC by k [np.float64(157.8), np.float64(130.8), np.float64(57.2), np.float64(49.0), np.float64(37.3), np.float64(26.4), np.float64(8.8)]\ntbilrate BIC best k = 5 ['1966Q1', '1978Q3', '1982Q3', '1991Q1', '2001Q3'] | BIC by k [np.float64(420.9), np.float64(377.2), np.float64(288.2), np.float64(237.1), np.float64(202.8), np.float64(169.0), np.float64(169.7)]\nPELT infl pen=10.6 ['1973Q1', '1981Q4']\nPELT infl pen=21.2 ['1973Q1', '1981Q4']\nPELT tbilrate pen=10.6 ['1966Q1', '1978Q3', '1982Q3', '1991Q1', '2001Q3']\nPELT tbilrate pen=21.2 ['1978Q3', '1985Q2', '2001Q2']\nPELT unemp pen=10.6 ['1974Q4', '1987Q2']\nPELT unemp pen=21.2 ['1974Q4', '1987Q2']\nBinseg infl ['1965Q3', '1973Q1', '1981Q4'] BottomUp ['1965Q3', '1971Q4', '1984Q2']\nBinseg tbilrate ['1978Q1', '1985Q3', '1991Q4'] BottomUp ['1978Q1', '1984Q2', '1990Q3']\nmulti BIC best 6 ['1965Q4', '1973Q3', '1978Q3', '1982Q3', '1986Q3', '1997Q1'] [np.float64(15.9), np.float64(-43.2), np.float64(-233.9), np.float64(-293.4), np.float64(-337.8), np.float64(-373.6), np.float64(-405.3)]\n\n[Notebook session — variables in memory: df: DataFrame 203x13, calc: Series 203, load(), d: DataFrame 202x15, kpss(), zivot_andrews(), DFGLS(), PhillipsPerron(), rows: list[15], v: str, x: Series 201, name: str, sub: Series 103, adf: tuple[6], k: tuple[4], dg: float64, pp: float64, st: DataFrame 15x8, za: tuple[5], code: str, s: str, ms: str, uc: str, MarkovRegression(), y: ndarray (201,), X: ndarray (201, 1), mod: MarkovRegression, t: float, r: MarkovRegressionResultsWrapper, CostLinear(), n: int, idx: PeriodIndex, MIN: int, bic_select(), res: dict[4], sig: ndarray (202, 1), o: list[7], kbest: tuple[4], sig_: ndarray (202, 1), pen: float64]","result":null,"figures":[],"files":[]}},{"tool":"execute_custom_python_analysis","result":{"success":true,"output":"Phillips ADL(2) {'supW_all': np.float64(44.379), 'date_all': '1981Q4', 'p_all': np.float64(0.004), 'supW_slope': np.float64(4.877), 'date_slope': '1982Q3', 'p_slope': np.float64(0.433)}\nTaylor (partial adj.) {'supW_all': np.float64(47.908), 'date_all': '1980Q3', 'p_all': np.float64(0.076), 'supW_slope': np.float64(10.465), 'date_slope': '1980Q3', 'p_slope': np.float64(0.766)}\n\n[Notebook session — variables in memory: df: DataFrame 203x13, calc: Series 203, load(), d: DataFrame 202x16, kpss(), zivot_andrews(), DFGLS(), PhillipsPerron(), rows: list[15], v: str, x: Series 201, name: str, sub: Series 103, adf: tuple[6], k: tuple[4], dg: float64, pp: float64, st: DataFrame 15x8, za: tuple[5], code: str, s: str, ms: str, uc: str, MarkovRegression(), y: ndarray (201,), X: ndarray (199, 4), mod: MarkovRegression, t: float, r: MarkovRegressionResultsWrapper, CostLinear(), n: int, idx: PeriodIndex, MIN: int, bic_select(), res: dict[4], sig: ndarray (202, 1), o: list[7], kbest: tuple[4], sig_: ndarray (202, 1), pen: float64]","result":null,"figures":[],"files":[]}},{"tool":"execute_custom_python_analysis","result":{"success":true,"output":"PC intercept+AR break: supW=43.42 at 1981Q4 p=0.002\nTR intercept+rho break: supW=27.35 at 1980Q3 p=0.178\nPC Chow 1973Q1 F=1.11 p=0.3530 (asymptotic F, no bootstrap)\nPC Chow 1979Q3 F=3.01 p=0.0194 (asymptotic F, no bootstrap)\nPC Chow 1984Q1 F=6.45 p=0.0001 (asymptotic F, no bootstrap)\nTR Chow 1973Q1 F=1.41 p=0.2331 (asymptotic F, no bootstrap)\nTR Chow 1979Q3 F=2.33 p=0.0573 (asymptotic F, no bootstrap)\nTR Chow 1984Q1 F=1.42 p=0.2283 (asymptotic F, no bootstrap)\nPC OLS-CUSUM stat 1.014 p=0.2558\nTR OLS-CUSUM stat 0.900 p=0.3925\nPC BIC [np.float64(365.3), np.float64(350.2), np.float64(355.4), np.float64(367.9), np.float64(379.7)] -> k= 1 ['1981Q4'] | LWZ k= 0 []\n   k=1 ['1981Q4']\n   k=2 ['1973Q1', '1981Q4']\n   k=3 ['1973Q1', '1981Q4', '1991Q1']\nTR BIC [np.float64(-52.6), np.float64(-70.7), np.float64(-112.4), np.float64(-139.2), np.float64(-128.7)] -> k= 3 ['1976Q4', '1980Q4', '1985Q1'] | LWZ k= 2 ['1980Q4', '1985Q1']\n   k=1 ['1980Q3']\n   k=2 ['1980Q4', '1985Q1']\n   k=3 ['1976Q4', '1980Q4', '1985Q1']\n\n[Notebook session — variables in memory: df: DataFrame 203x13, calc: Series 203, load(), d: DataFrame 202x16, kpss(), zivot_andrews(), DFGLS(), PhillipsPerron(), rows: list[15], v: str, x: Series 201, name: str, sub: Series 103, adf: tuple[6], k: tuple[4], dg: float64, pp: float64, st: DataFrame 15x8, za: tuple[5], code: str, s: str, ms: str, uc: str, MarkovRegression(), y: ndarray (201,), X: ndarray (200, 4), mod: MarkovRegression, t: int, r: tuple[5], CostLinear(), n: int, idx: PeriodIndex, MIN: int, bic_select(), res: dict[4], sig: ndarray (202, 1), o: list[5], kbest: tuple[4], sig_: ndarray (202, 1), pen: float64]","result":null,"figures":[],"files":[]}},{"tool":"execute_custom_python_analysis","result":{"success":true,"output":"seg    n  slope_u   se_u  persist  lr_slope  se_lr  resid_sd\n0  1959Q4–2009Q3  200    0.012  0.096    0.754     0.049  0.383     2.388 \n\n             seg    n  slope_u   se_u  persist  lr_slope  se_lr  resid_sd\n0  1959Q4–1981Q3   88   -0.120  0.100    0.916    -1.427  1.791     2.017\n1  1981Q4–2009Q3  112    0.104  0.101    0.118     0.117  0.115     2.290 \n\n             seg    n  slope_u   se_u  persist  lr_slope  se_lr  resid_sd\n0  1959Q4–1972Q4   53   -0.547  0.169    0.618    -1.430  0.352     1.245\n1  1973Q1–1981Q3   35   -1.075  0.212    0.424    -1.868  0.482     2.369\n2  1981Q4–2009Q3  112    0.104  0.101    0.118     0.117  0.115     2.290 \n\n             seg    n    rho  b_pi_sr  c_u_sr  lr_pi  se_lr_pi   lr_u  resid_sd\n0  1960Q1–2009Q3  199  0.886    0.116  -0.055   1.02     0.252 -0.479     0.839 \n\n             seg   n    rho  b_pi_sr  c_u_sr  lr_pi  se_lr_pi   lr_u  resid_sd\n0  1960Q1–1980Q3  83  0.243    0.505  -0.369  0.668     0.061 -0.487     0.740\n1  1980Q4–1984Q4  17  0.008    0.799  -0.228  0.806     0.057 -0.230     1.027\n2  1985Q1–2009Q3  99  0.956    0.054  -0.054  1.245     1.045 -1.232     0.470 \n\nvariance breaks k=1 ['2004Q4']\nvariance breaks k=2 ['1987Q2', '2004Q4']\npre/post 1984 infl sd 3.76 1.75\n\n[Notebook session — variables in memory: df: DataFrame 203x13, calc: Series 203, load(), d: DataFrame 202x16, kpss(), zivot_andrews(), DFGLS(), PhillipsPerron(), rows: list[15], v: str, x: Series 201, name: str, sub: Series 103, adf: tuple[6], k: int, dg: float64, pp: float64, st: DataFrame 15x8, za: tuple[5], code: str, s: str, ms: str, uc: str, MarkovRegression(), y: ndarray (201,), X: ndarray (200, 4), mod: MarkovRegression, t: DataFrame 3x9, r: RegressionResultsWrapper, CostLinear(), n: int, idx: PeriodIndex, MIN: int, bic_select(), res: dict[4], sig: ndarray (202, 1), o: list[5], kbest: tuple[4], sig_: ndarray (202, 1), pen: float64]","result":null,"figures":[],"files":[]}},{"tool":"research_runs","result":{"runs":[{"run_id":"8125167d-cec8-434f-b799-c5365fcf2a6c","label":"Markov-switching expanding","code":"code/fc_ms.py","status":"done","exit_code":0,"seconds":38,"progress":1,"error":null,"logs":"runs/8125167d-cec8-434f-b799-c5365fcf2a6c/stdout.log, runs/8125167d-cec8-434f-b799-c5365fcf2a6c/stderr.log"},{"run_id":"a0e1b107-631d-489d-a128-c8f32694c432","label":"Markov-switching rolling 60","code":"code/fc_ms.py","status":"done","exit_code":0,"seconds":45,"progress":1,"error":null,"logs":"runs/a0e1b107-631d-489d-a128-c8f32694c432/stdout.log, runs/a0e1b107-631d-489d-a128-c8f32694c432/stderr.log"},{"run_id":"e887c62f-bb1d-45af-9fcd-47088bd65bbf","label":"UCSV particle filter gamma 0.2","code":"code/fc_ucsv.py","status":"done","exit_code":0,"seconds":2,"progress":1,"error":null,"logs":"runs/e887c62f-bb1d-45af-9fcd-47088bd65bbf/stdout.log, runs/e887c62f-bb1d-45af-9fcd-47088bd65bbf/stderr.log"},{"run_id":"a2531baa-439f-4776-a522-baced94a1bf1","label":"UCSV particle filter gamma 0.1","code":"code/fc_ucsv.py","status":"done","exit_code":0,"seconds":2,"progress":1,"error":null,"logs":"runs/a2531baa-439f-4776-a522-baced94a1bf1/stdout.log, runs/a2531baa-439f-4776-a522-baced94a1bf1/stderr.log"}]}},{"tool":"execute_custom_python_analysis","result":{"success":true,"output":"linear_expanding (3123, 6) 0\nlinear_rolling40 (3123, 6) 0\nlinear_rolling60 (3123, 6) 0\nms_expanding (466, 6) 0\nms_rolling60 (466, 6) 19\nucsv_g0.1 (347, 6) 0\nucsv_g0.2 (347, 6) 0\nms_expanding fit failures: {'MS_AR': 0, 'MS_PC': 0} NaN forecasts: 0\nmodel  MS_AR  MS_PC\nh                  \n1      2.711  2.834\n4      2.014    NaN\n8      1.799    NaN\n\nms_rolling60 fit failures: {'MS_AR': 2, 'MS_PC': 16} NaN forecasts: 19\nmodel  MS_AR  MS_PC\nh                  \n1      2.619  2.769\n4      1.859    NaN\n8      1.801    NaN\n\n[Notebook session — variables in memory: df: DataFrame 203x13, calc: Series 203, load(), d: DataFrame 202x16, kpss(), zivot_andrews(), DFGLS(), PhillipsPerron(), rows: list[15], v: str, x: DataFrame 347x6, name: str, sub: Series 103, adf: tuple[6], k: int, dg: float64, pp: float64, st: DataFrame 15x8, za: tuple[5], code: str, s: str, ms: str, uc: str, MarkovRegression(), y: ndarray (201,), X: ndarray (200, 4), mod: MarkovRegression, t: DataFrame 3x9, r: str, CostLinear(), n: int, idx: PeriodIndex, MIN: int, bic_select(), res: dict[4], sig: ndarray (202, 1), o: list[5], kbest: tuple[4], sig_: ndarray (202, 1), pen: float64]","result":null,"figures":[],"files":[]}},{"tool":"research_run","result":{"run_id":"8f82b8f5-d691-4ba0-b8fa-cb325149fc11","label":"Markov-switching rolling 40","code":"code/fc_ms.py","status":"running","exit_code":null,"seconds":0,"progress":null,"error":null,"logs":"runs/8f82b8f5-d691-4ba0-b8fa-cb325149fc11/stdout.log, runs/8f82b8f5-d691-4ba0-b8fa-cb325149fc11/stderr.log","note":"Started in the background. Use research_runs (wait / status / logs) to follow it; start other runs meanwhile if they are independent."}},{"tool":"execute_custom_python_analysis","result":{"success":true,"output":"{1: 119, 4: 116, 8: 112} 228\n\n[Notebook session — variables in memory: df: DataFrame 203x13, calc: Series 203, load(), d: DataFrame 202x16, kpss(), zivot_andrews(), DFGLS(), PhillipsPerron(), rows: list[15], v: str, x: DataFrame 347x6, name: str, sub: Series 103, adf: tuple[6], k: int, dg: float64, pp: float64, st: DataFrame 15x8, za: tuple[5], code: str, s: str, ms: str, uc: str, MarkovRegression(), y: ndarray (201,), X: ndarray (200, 4), mod: MarkovRegression, t: DataFrame 3x9, r: NoneType, CostLinear(), n: int, idx: PeriodIndex, MIN: int, bic_select(), res: dict[4], sig: ndarray (202, 1), o: list[5], kbest: tuple[4], sig_: ndarray (202, 1), pen: float64]","result":null,"figures":[],"files":[]}},{"tool":"execute_custom_python_analysis","result":{"success":true,"output":"EXPANDING RMSFE\n               1      4      8\nAR        2.595  1.823  1.629\nComb      2.545  1.862  1.824\nIMA11     2.567  1.821  1.740\nMS_AR     2.711  2.014  1.799\nMS_PC     2.834    NaN    NaN\nPC_gap1   2.626  1.841  1.733\nPC_u      2.730  1.874  1.883\nPC_ut     2.843  2.393  2.662\nRW1       3.020  2.411  2.226\nRW4       2.591  1.842  1.779\nRidge     2.683  2.295  2.392\nUCSV      2.466  1.688  1.701\nUCSV_g01  2.495  1.743  1.719\nVAR3      2.808  2.296  2.476\n\nrelative to AR\n               1      4      8\nAR        1.000  1.000  1.000\nComb      0.980  1.022  1.120\nIMA11     0.989  0.999  1.068\nMS_AR     1.045  1.105  1.104\nMS_PC     1.092    NaN    NaN\nPC_gap1   1.012  1.010  1.064\nPC_u      1.052  1.028  1.156\nPC_ut     1.095  1.313  1.634\nRW1       1.164  1.323  1.366\nRW4       0.998  1.010  1.092\nRidge     1.034  1.259  1.468\nUCSV      0.950  0.926  1.044\nUCSV_g01  0.961  0.956  1.055\nVAR3      1.082  1.260  1.520\n\nROLLING60 RMSFE\n               1      4      8\nAR        2.581  1.710  1.735\nComb      2.582  1.873  1.939\nIMA11     2.469  1.692  1.702\nMS_AR     2.619  1.859  1.801\nMS_PC     2.769    NaN    NaN\nPC_gap1   2.660  1.832  2.233\nPC_u      2.708  1.912  2.411\nPC_ut     2.922  2.560  2.898\nRW1       3.020  2.411  2.226\nRW4       2.591  1.842  1.779\nRidge     2.568  2.380  2.699\nUCSV      2.466  1.688  1.701\nUCSV_g01  2.495  1.743  1.719\nVAR3      2.953  2.377  2.690\n\nrel\n               1      4      8\nAR        1.000  1.000  1.000\nComb      1.000  1.095  1.118\nIMA11     0.956  0.990  0.981\nMS_AR     1.015  1.087  1.038\nMS_PC     1.073    NaN    NaN\nPC_gap1   1.030  1.072  1.287\nPC_u      1.049  1.118  1.390\nPC_ut     1.132  1.497  1.670\nRW1       1.170  1.410  1.283\nRW4       1.004  1.077  1.025\nRidge     0.995  1.392  1.556\nUCSV      0.955  0.987  0.980\nUCSV_g01  0.967  1.019  0.991\nVAR3      1.144  1.390  1.550\n\n[Notebook session — variables in memory: df: DataFrame 203x13, calc: Series 203, load(), d: DataFrame 202x16, kpss(), zivot_andrews(), DFGLS(), PhillipsPerron(), rows: list[15], v: str, x: DataFrame 347x6, name: str, sub: Series 103, adf: tuple[6], k: int, dg: float64, pp: float64, st: DataFrame 15x8, za: tuple[5], code: str, s: str, ms: str, uc: str, MarkovRegression(), y: ndarray (201,), X: ndarray (200, 4), mod: MarkovRegression, t: DataFrame 3x9, r: NoneType, CostLinear(), n: int, idx: PeriodIndex, MIN: int, bic_select(), res: dict[4], sig: ndarray (202, 1), o: list[5], kbest: tuple[4], sig_: ndarray (202, 1), pen: float64]","result":null,"figures":[],"files":[]}},{"tool":"execute_custom_python_analysis","result":{"success":true,"output":"h     model    n  rel_rmsfe  DM_stat   DM_p  CW_stat   CW_p\n0   1       RW4  119      0.998   -0.054  0.957      NaN    NaN\n1   1     IMA11  119      0.989   -0.392  0.696      NaN    NaN\n2   1      UCSV  119      0.950   -1.362  0.176      NaN    NaN\n3   1  UCSV_g01  119      0.961   -1.255  0.212      NaN    NaN\n4   1     MS_AR  119      1.045    0.751  0.454    0.959  0.169\n5   1     MS_PC  119      1.092    1.310  0.193    1.210  0.113\n6   1      PC_u  119      1.052    0.872  0.385    0.167  0.434\n7   1   PC_gap1  119      1.012    0.687  0.493    0.056  0.478\n8   1     PC_ut  119      1.095    1.407  0.162    0.186  0.426\n9   1      VAR3  119      1.082    1.250  0.214    0.078  0.469\n10  1     Ridge  119      1.034    0.808  0.421    1.742  0.041\n11  1      Comb  119      0.980   -0.780  0.437      NaN    NaN\n12  4       RW4  116      1.010    0.203  0.840      NaN    NaN\n13  4     IMA11  116      0.999   -0.035  0.972      NaN    NaN\n14  4      UCSV  116      0.926   -1.146  0.254      NaN    NaN\n15  4  UCSV_g01  116      0.956   -0.880  0.381      NaN    NaN\n16  4     MS_AR  116      1.105    1.443  0.152    1.077  0.141\n17  4      PC_u  116      1.028    0.665  0.507    0.386  0.350\n18  4   PC_gap1  116      1.010    0.632  0.529    0.030  0.488\n19  4     PC_ut  116      1.313    1.765  0.080   -1.150  0.875\n20  4      VAR3  116      1.260    1.770  0.079   -0.687  0.754\n21  4     Ridge  116      1.259    1.295  0.198    0.557  0.289\n22  4      Comb  116      1.022    0.478  0.634      NaN    NaN\n23  8       RW4  112      1.092    0.607  0.545      NaN    NaN\n24  8     IMA11  112      1.068    0.552  0.582      NaN    NaN\n25  8      UCSV  112      1.044    0.341  0.734      NaN    NaN\n26  8  UCSV_g01  112      1.055    0.420  0.675      NaN    NaN\n27  8     MS_AR  112      1.104    0.941  0.349    1.250  0.106\n28  8      PC_u  112      1.156    1.699  0.092   -1.480  0.931\n29  8   PC_gap1  112      1.064    1.708  0.091   -1.781  0.963\n30  8     PC_ut  112      1.634    1.369  0.174   -1.346  0.911\n31  8      VAR3  112      1.520    1.416  0.160   -0.405  0.657\n32  8     Ridge  112      1.468    1.245  0.216   -0.395  0.654\n33  8      Comb  112      1.120    0.827  0.410      NaN    NaN\n    h  model    n  rel_rmsfe  DM_stat   DM_p  CW_stat  CW_p\n0   1     AR  119      1.053    1.362  0.176      NaN   NaN\n1   1    RW4  119      1.051    1.264  0.209      NaN   NaN\n2   1  IMA11  119      1.041    1.364  0.175      NaN   NaN\n3   1  MS_AR  119      1.099    1.546  0.125      NaN   NaN\n4   1   PC_u  119      1.107    1.222  0.224      NaN   NaN\n5   1  PC_ut  119      1.153    1.640  0.104      NaN   NaN\n6   1   VAR3  119      1.139    1.590  0.115      NaN   NaN\n7   1  Ridge  119      1.088    1.564  0.121      NaN   NaN\n8   1   Comb  119      1.032    0.811  0.419      NaN   NaN\n9   4     AR  116      1.080    1.146  0.254      NaN   NaN\n10  4    RW4  116      1.091    1.502  0.136      NaN   NaN\n11  4  IMA11  116      1.079    1.465  0.146      NaN   NaN\n12  4  MS_AR  116      1.193    1.502  0.136      NaN   NaN\n13  4   PC_u  116      1.110    1.771  0.079      NaN   NaN\n14  4  PC_ut  116      1.418    2.251  0.026      NaN   NaN\n15  4   VAR3  116      1.360    2.252  0.026      NaN   NaN\n16  4  Ridge  116      1.359    1.745  0.084      NaN   NaN\n17  4   Comb  116      1.103    1.527  0.129      NaN   NaN\n18  8     AR  112      0.958   -0.341  0.734      NaN   NaN\n19  8    RW4  112      1.046    1.271  0.207      NaN   NaN\n20  8  IMA11  112      1.023    1.168  0.245      NaN   NaN\n21  8  MS_AR  112      1.058    0.312  0.755      NaN   NaN\n22  8   PC_u  112      1.107    0.943  0.348      NaN   NaN\n23  8  PC_ut  112      1.565    1.562  0.121      NaN   NaN\n24  8   VAR3  112      1.456    1.773  0.079      NaN   NaN\n25  8  Ridge  112      1.407    1.453  0.149      NaN   NaN\n26  8   Comb  112      1.073    1.280  0.203      NaN   NaN\n\n[Notebook session — variables in memory: df: DataFrame 203x13, calc: Series 203, load(), d: DataFrame 202x16, kpss(), zivot_andrews(), DFGLS(), PhillipsPerron(), rows: list[15], v: str, x: DataFrame 347x6, name: str, sub: Series 103, adf: tuple[6], k: int, dg: float64, pp: float64, st: DataFrame 15x8, za: tuple[5], code: str, s: str, ms: str, uc: str, MarkovRegression(), y: ndarray (201,), X: ndarray (200, 4), mod: MarkovRegression, t: DataFrame 3x9, r: NoneType, CostLinear(), n: int, idx: PeriodIndex, MIN: int, bic_select(), res: dict[4], sig: ndarray (202, 1), o: list[5], kbest: tuple[4], sig_: ndarray (202, 1), pen: float64]","result":null,"figures":[],"files":[]}},{"tool":"execute_custom_python_analysis","result":{"success":true,"output":"h                           1      4      8\nperiod           model                     \n1980-1989        Comb   1.007  1.094  1.235\n                 IMA11  1.001  1.056  1.206\n                 MS_AR  1.066  1.047  1.090\n                 PC_u   1.008  1.096  1.238\n                 PC_ut  1.116  1.614  1.878\n                 RW4    0.974  1.066  1.249\n                 Ridge  1.029  1.576  1.715\n                 UCSV   1.015  1.060  1.207\n                 VAR3   1.143  1.519  1.769\n1990-2007        Comb   1.003  0.940  0.923\n                 IMA11  1.061  0.937  0.819\n                 MS_AR  1.195  1.132  1.041\n                 PC_u   0.998  1.037  1.032\n                 PC_ut  1.011  1.077  1.210\n                 RW4    1.056  0.960  0.805\n                 Ridge  1.043  0.956  1.020\n                 UCSV   1.016  0.838  0.754\n                 VAR3   1.008  1.028  1.065\nall              Comb   0.980  1.022  1.120\n                 IMA11  0.989  0.999  1.068\n                 MS_AR  1.045  1.105  1.104\n                 PC_u   1.052  1.028  1.156\n                 PC_ut  1.095  1.313  1.634\n                 RW4    0.998  1.010  1.092\n                 Ridge  1.034  1.259  1.468\n                 UCSV   0.950  0.926  1.044\n                 VAR3   1.082  1.260  1.520\nex-GFC (<2008Q3) Comb   1.005  1.036  1.132\n                 IMA11  1.026  1.011  1.083\n                 MS_AR  1.119  1.081  1.072\n                 PC_u   1.004  1.073  1.168\n                 PC_ut  1.076  1.427  1.668\n                 RW4    1.008  1.025  1.110\n                 Ridge  1.035  1.367  1.502\n                 UCSV   1.015  0.979  1.067\n                 VAR3   1.091  1.348  1.553\n                      n  AR_rmsfe\nperiod           h               \n1980-1989        1   40      2.79\n                 4   40      2.03\n                 8   40      2.12\n1990-2007        1   74      1.67\n                 4   71      1.23\n                 8   67      1.23\nall              1  119      2.60\n                 4  116      1.82\n                 8  112      1.63\nex-GFC (<2008Q3) 1  114      2.13\n                 4  111      1.57\n                 8  107      1.62\n              period  h    n  AR_rmsfe  model    rel   DM_p\n14         1980-1989  4   40     2.035  PC_ut  1.614  0.089\n15         1980-1989  4   40     2.035   VAR3  1.519  0.089\n22         1980-1989  8   40     2.116   PC_u  1.238  0.075\n1          1990-2007  1   74     1.669  IMA11  1.061  0.038\n3          1990-2007  1   74     1.669  MS_AR  1.195  0.061\n11         1990-2007  4   71     1.229   UCSV  0.838  0.094\n17         1990-2007  4   71     1.229   Comb  0.940  0.083\n18         1990-2007  8   67     1.226    RW4  0.805  0.075\n19         1990-2007  8   67     1.226  IMA11  0.819  0.075\n20         1990-2007  8   67     1.226   UCSV  0.754  0.022\n26         1990-2007  8   67     1.226   Comb  0.923  0.096\n3   ex-GFC (<2008Q3)  1  114     2.132  MS_AR  1.119  0.047\n14  ex-GFC (<2008Q3)  4  111     1.568  PC_ut  1.427  0.075\n15  ex-GFC (<2008Q3)  4  111     1.568   VAR3  1.348  0.084\n22  ex-GFC (<2008Q3)  8  107     1.617   PC_u  1.168  0.084\n14               all  4  116     1.823  PC_ut  1.313  0.080\n15               all  4  116     1.823   VAR3  1.260  0.079\n22               all  8  112     1.629   PC_u  1.156  0.092\n\n[Notebook session — variables in memory: df: DataFrame 203x13, calc: Series 203, load(), d: DataFrame 202x16, kpss(), zivot_andrews(), DFGLS(), PhillipsPerron(), rows: list[27], v: str, x: DataFrame 112x15, name: str, sub: dict[4], adf: tuple[6], k: int, dg: float64, pp: float64, st: DataFrame 15x8, za: tuple[5], code: str, s: str, ms: str, uc: str, MarkovRegression(), y: ndarray (201,), X: ndarray (200, 4), mod: MarkovRegression, t: DataFrame 3x9, r: NoneType, CostLinear(), n: int, idx: PeriodIndex, MIN: int, bic_select(), res: dict[4], sig: ndarray (202, 1), o: list[5], kbest: tuple[4], sig_: ndarray (202, 1), pen: float64]","result":null,"figures":[],"files":[]}},{"tool":"execute_custom_python_analysis","result":{"success":true,"output":"VAR whiteness p 0.0\nVAR normality p 0.0\nGranger p: {'unemp': np.float64(0.9756), 'tbilrate': np.float64(0.0064)}\n                             Regime 0 parameters                              \n==============================================================================\n                 coef    std err          z      P>|z|      [0.025      0.975]\n------------------------------------------------------------------------------\nconst          2.8136      0.616      4.568      0.000       1.606       4.021\nsigma2        13.6402      2.461      5.543      0.000       8.817      18.463\n==============================================================================\n                   model    n      sd   LB4_p   LB8_p  LB_sq4_p  ARCH4_p    JB_p    skew     kurt\n0                  AR(3)  199  2.2635  0.9513  0.7283    0.0000   0.0000  0.0000 -0.7413   7.4065\n1            PC ADL(2)+u  200  2.3639  0.0038  0.0101    0.0000   0.0000  0.0000 -0.6461   6.7249\n2              Taylor PA  199  0.8307  0.0056  0.0002    0.0000   0.0000  0.0000 -1.3114  17.5965\n3         VAR(2) eq infl  200  2.2983  0.0062  0.0058    0.0000   0.0000  0.0000 -0.6716   6.5801\n4        VAR(2) eq unemp  200  0.2380  0.5510  0.0415    0.0000   0.0000  0.0002  0.4162   4.1793\n5     VAR(2) eq tbilrate  200  0.8374  0.0001  0.0000    0.0000   0.0000  0.0000 -0.7569  17.7819\n6  MS-AR(1) (std. resid)  201  0.9198  0.0000  0.0000    0.2339   0.1464  0.0175 -0.4870   3.1311\nexpected durations [16.11928509 29.65907525]\n\n[Notebook session — variables in memory: df: DataFrame 203x13, calc: Series 203, load(), d: DataFrame 202x16, kpss(), zivot_andrews(), DFGLS(), PhillipsPerron(), rows: list[27], v: VARResultsWrapper, x: DataFrame 112x15, name: str, sub: dict[4], adf: tuple[6], k: int, dg: float64, pp: float64, st: DataFrame 15x8, za: tuple[5], code: str, s: str, ms: str, uc: str, MarkovRegression(), y: ndarray (201,), X: ndarray (200, 4), mod: MarkovRegression, t: DataFrame 3x9, r: NoneType, CostLinear(), n: int, idx: PeriodIndex, MIN: int, bic_select(), res: dict[4], sig: ndarray (202, 1), o: list[5], kbest: tuple[4], sig_: ndarray (202, 1), pen: float64]","result":null,"figures":[],"files":[]}},{"tool":"execute_custom_python_analysis","result":{"success":true,"output":"{'p[0->0]': np.float64(0.938), 'p[1->0]': np.float64(0.034), 'const[0]': np.float64(2.814), 'const[1]': np.float64(1.376), 'x1[1]': np.float64(0.529), 'sigma2[0]': np.float64(13.64), 'sigma2[1]': np.float64(1.607)}\n                            model    n      sd   LB4_p  LB8_p  LB_sq4_p  ARCH4_p  JB_p    skew     kurt\n5              VAR(2) eq tbilrate  200  0.8374  0.0001    0.0    0.0000   0.0000   0.0 -0.7569  17.7819\n6  MS-AR(1) (std. one-step resid)  200  1.0169  0.0000    0.0    0.6974   0.6882   0.0 -0.0914   6.4796\n\n[Notebook session — variables in memory: df: DataFrame 203x13, calc: Series 203, load(), d: DataFrame 202x16, kpss(), zivot_andrews(), DFGLS(), PhillipsPerron(), rows: list[27], v: VARResultsWrapper, x: DataFrame 112x15, name: str, sub: dict[4], adf: tuple[6], k: int, dg: float64, pp: float64, st: DataFrame 15x8, za: tuple[5], code: str, s: str, ms: str, uc: str, MarkovRegression(), y: ndarray (201,), X: ndarray (200, 4), mod: MarkovRegression, t: DataFrame 3x9, r: NoneType, CostLinear(), n: int, idx: PeriodIndex, MIN: int, bic_select(), res: dict[4], sig: ndarray (202, 1), o: list[5], kbest: tuple[4], sig_: ndarray (202, 1), pen: float64]","result":null,"figures":[],"files":[]}},{"tool":"execute_custom_python_analysis","result":{"success":true,"output":"2120\n\n[Notebook session — variables in memory: df: DataFrame 203x13, calc: Series 203, load(), d: DataFrame 202x16, kpss(), zivot_andrews(), DFGLS(), PhillipsPerron(), rows: list[27], v: VARResultsWrapper, x: DataFrame 112x15, name: str, sub: dict[4], adf: tuple[6], k: int, dg: float64, pp: float64, st: DataFrame 15x8, za: tuple[5], code: str, s: str, ms: str, uc: str, MarkovRegression(), y: ndarray (201,), X: ndarray (200, 4), mod: MarkovRegression, t: DataFrame 3x9, r: NoneType, CostLinear(), n: int, idx: PeriodIndex, MIN: int, bic_select(), res: dict[4], sig: ndarray (202, 1), o: list[5], kbest: tuple[4], sig_: ndarray (202, 1), pen: float64]","result":null,"figures":[],"files":[]}},{"tool":"research_run","result":{"run_id":"0b4b4569-3c89-4bdf-b7c9-ff5a75c1afff","label":"MS-AR(3) expanding","code":"code/fc_ms3.py","status":"running","exit_code":null,"seconds":0,"progress":null,"error":null,"logs":"runs/0b4b4569-3c89-4bdf-b7c9-ff5a75c1afff/stdout.log, runs/0b4b4569-3c89-4bdf-b7c9-ff5a75c1afff/stderr.log","note":"Started in the background. Use research_runs (wait / status / logs) to follow it; start other runs meanwhile if they are independent."}},{"tool":"research_run","result":{"run_id":"ce22ffca-f30e-4f4b-bab8-eed096b6ab28","label":"MS-AR(3) rolling 60","code":"code/fc_ms3.py","status":"running","exit_code":null,"seconds":0,"progress":null,"error":null,"logs":"runs/ce22ffca-f30e-4f4b-bab8-eed096b6ab28/stdout.log, runs/ce22ffca-f30e-4f4b-bab8-eed096b6ab28/stderr.log","note":"Started in the background. Use research_runs (wait / status / logs) to follow it; start other runs meanwhile if they are independent."}},{"tool":"log_experiment","result":{"experiment_id":"880c79ea-a091-43c4-bc0e-d38f5f81f624","number":33,"status":"ok","recorded":true}},{"tool":"log_experiment","result":{"experiment_id":"28b5d2e8-2c04-44db-b96e-e32e35a0cb42","number":34,"status":"ok","recorded":true}},{"tool":"log_experiment","result":{"experiment_id":"95294c79-17d5-4d34-b387-95e51f13fbea","number":35,"status":"ok","recorded":true}},{"tool":"execute_custom_python_analysis","result":{"success":true,"output":"1 {'PC_du': np.float64(1.02), 'PC_ugapMA': np.float64(1.015), 'PC_rr': np.float64(1.049), 'AR4': np.float64(1.01), 'PC_u': np.float64(1.052)}\n   DM p PC_du 0.674\n   DM p PC_ugapMA 0.464\n   DM p PC_rr 0.073\n\n4 {'PC_du': np.float64(1.014), 'PC_ugapMA': np.float64(1.006), 'PC_rr': np.float64(1.188), 'AR4': np.float64(0.998), 'PC_u': np.float64(1.028)}\n   DM p PC_du 0.834\n   DM p PC_ugapMA 0.796\n   DM p PC_rr 0.077\n\n8 {'PC_du': np.float64(1.103), 'PC_ugapMA': np.float64(1.014), 'PC_rr': np.float64(1.185), 'AR4': np.float64(1.01), 'PC_u': np.float64(1.156)}\n   DM p PC_du 0.247\n   DM p PC_ugapMA 0.63\n   DM p PC_rr 0.154\n\n[Notebook session — variables in memory: df: DataFrame 203x13, calc: Series 203, load(), d: DataFrame 202x16, kpss(), zivot_andrews(), DFGLS(), PhillipsPerron(), rows: list[1388], v: VARResultsWrapper, x: DataFrame 112x19, name: str, sub: dict[4], adf: tuple[6], k: int, dg: float64, pp: float64, st: DataFrame 15x8, za: tuple[5], code: str, s: str, ms: str, uc: str, MarkovRegression(), y: ndarray (202,), X: ndarray (176, 5), mod: MarkovRegression, t: DataFrame 3x9, r: NoneType, CostLinear(), n: int, idx: PeriodIndex, MIN: int, bic_select(), res: dict[4], sig: ndarray (202, 1), o: list[5], kbest: tuple[4], sig_: ndarray (202, 1), pen: float64]","result":null,"figures":[],"files":[]}},{"tool":"research_runs","result":{"runs":[{"run_id":"0b4b4569-3c89-4bdf-b7c9-ff5a75c1afff","label":"MS-AR(3) expanding","code":"code/fc_ms3.py","status":"done","exit_code":0,"seconds":32,"progress":1,"error":null,"logs":"runs/0b4b4569-3c89-4bdf-b7c9-ff5a75c1afff/stdout.log, runs/0b4b4569-3c89-4bdf-b7c9-ff5a75c1afff/stderr.log","stdout_tail":"ms3_expanding fit failures 0 NaN 0\nh\n1    2.643\n4    1.844\n8    1.739\ndtype: float64\n"},{"run_id":"ce22ffca-f30e-4f4b-bab8-eed096b6ab28","label":"MS-AR(3) rolling 60","code":"code/fc_ms3.py","status":"done","exit_code":0,"seconds":42,"progress":1,"error":null,"logs":"runs/ce22ffca-f30e-4f4b-bab8-eed096b6ab28/stdout.log, runs/ce22ffca-f30e-4f4b-bab8-eed096b6ab28/stderr.log","stdout_tail":"ms3_rolling60 fit failures 13 NaN 13\nh\n1    2.457\n4    1.875\n8    1.772\ndtype: float64\n","stderr_tail":"ng/markov_switching.py:1292: EstimationWarning: Invalid regime transition probabilities estimated in EM iteration; probabilities have been re-scaled to continue estimation.\n  warnings.warn('Invalid regime transition probabilities'\n/Users/simon-pierreboucher/dev/vquant-next/.venv/lib/python3.14/site-packages/statsmodels/tsa/regime_switching/markov_switching.py:1292: EstimationWarning: Invalid regime transition probabilities estimated in EM iteration; probabilities have been re-scaled to continue estimation.\n  warnings.warn('Invalid regime transition probabilities'\n/Users/simon-pierreboucher/dev/vquant-next/.venv/lib/python3.14/site-packages/statsmodels/tsa/regime_switching/markov_switching.py:1292: EstimationWarning: Invalid regime transition probabilities estimated in EM iteration; probabilities have been re-scaled to continue estimation.\n  warnings.warn('Invalid regime transition probabilities'\n/Users/simon-pierreboucher/dev/vquant-next/.venv/lib/python3.14/site-packages/statsmodels/tsa/regime_switching/markov_switching.py:1292: EstimationWarning: Invalid regime transition probabilities estimated in EM iteration; probabilities have been re-scaled to continue estimation.\n  warnings.warn('Invalid regime transition probabilities'\n/Users/simon-pierreboucher/dev/vquant-next/.venv/lib/python3.14/site-packages/statsmodels/tsa/regime_switching/markov_switching.py:1292: EstimationWarning: Invalid regime transition probabilities estimated in EM iteration; probabilities have been re-scaled to continue estimation.\n  warnings.warn('Invalid regime transition probabilities'\n/Users/simon-pierreboucher/dev/vquant-next/.venv/lib/python3.14/site-packages/statsmodels/tsa/regime_switching/markov_switching.py:1292: EstimationWarning: Invalid regime transition probabilities estimated in EM iteration; probabilities have been re-scaled to continue estimation.\n  warnings.warn('Invalid regime transition probabilities'\n/Users/simon-pierreboucher/dev/vquant-next/.venv/lib/python3.14/site-packages/statsmodels/tsa/regime_switching/markov_switching.py:1292: EstimationWarning: Invalid regime transition probabilities estimated in EM iteration; probabilities have been re-scaled to continue estimation.\n  warnings.warn('Invalid regime transition probabilities'\n/Users/simon-pierreboucher/dev/vquant-next/.venv/lib/python3.14/site-packages/statsmodels/tsa/regime_switching/markov_switching.py:1292: EstimationWarning: Invalid regime transition probabilities estimated in EM iteration; probabilities have been re-scaled to continue estimation.\n  warnings.warn('Invalid regime transition probabilities'\n/Users/simon-pierreboucher/dev/vquant-next/.venv/lib/python3.14/site-packages/statsmodels/tsa/regime_switching/markov_switching.py:1292: EstimationWarning: Invalid regime transition probabilities estimated in EM iteration; probabilities have been re-scaled to continue estimation.\n  warnings.warn('Invalid regime transition probabilities'\n/Users/simon-pierreboucher/dev/vquant-next/.venv/lib/python3.14/site-packages/statsmodels/tsa/regime_switching/markov_switching.py:1292: EstimationWarning: Invalid regime transition probabilities estimated in EM iteration; probabilities have been re-scaled to continue estimation.\n  warnings.warn('Invalid regime transition probabilities'\n/Users/simon-pierreboucher/dev/vquant-next/.venv/lib/python3.14/site-packages/statsmodels/tsa/regime_switching/markov_switching.py:1292: EstimationWarning: Invalid regime transition probabilities estimated in EM iteration; probabilities have been re-scaled to continue estimation.\n  warnings.warn('Invalid regime transition probabilities'\n/Users/simon-pierreboucher/dev/vquant-next/.venv/lib/python3.14/site-packages/statsmodels/tsa/regime_switching/markov_switching.py:1292: EstimationWarning: Invalid regime transition probabilities estimated in EM iteration; probabilities have been re-scaled to continue estimation.\n  warnings.warn('Invalid regime transition probabilities'\n/Users/simon-pierreboucher/dev/vquant-next/.venv/lib/python3.14/site-packages/statsmodels/base/model.py:607: ConvergenceWarning: Maximum Likelihood optimization failed to converge. Check mle_retvals\n  warnings.warn(\"Maximum Likelihood optimization failed to \"\n/Users/simon-pierreboucher/dev/vquant-next/.venv/lib/python3.14/site-packages/statsmodels/base/model.py:607: ConvergenceWarning: Maximum Likelihood optimization failed to converge. Check mle_retvals\n  warnings.warn(\"Maximum Likelihood optimization failed to \"\n/Users/simon-pierreboucher/dev/vquant-next/.venv/lib/python3.14/site-packages/statsmodels/base/model.py:607: ConvergenceWarning: Maximum Likelihood optimization failed to converge. Check mle_retvals\n  warnings.warn(\"Maximum Likelihood optimization failed to \"\n/Users/simon-pierreboucher/dev/vquant-next/.venv/lib/python3.14/site-packages/statsmodels/base/model.py:607: ConvergenceWarning: Maximum Likelihood optimization failed to converge. Check mle_retvals\n  warnings.warn(\"Maximum Likelihood optimization failed to \""},{"run_id":"8f82b8f5-d691-4ba0-b8fa-cb325149fc11","label":"Markov-switching rolling 40","code":"code/fc_ms.py","status":"done","exit_code":0,"seconds":52,"progress":1,"error":null,"logs":"runs/8f82b8f5-d691-4ba0-b8fa-cb325149fc11/stdout.log, runs/8f82b8f5-d691-4ba0-b8fa-cb325149fc11/stderr.log","stdout_tail":"ms_rolling40 fit failures: {'MS_AR': 6, 'MS_PC': 28} NaN forecasts: 34\nmodel  MS_AR  MS_PC\nh                  \n1      2.674  2.496\n4      1.801    NaN\n8      1.861    NaN\n","stderr_tail":"v/vquant-next/.venv/lib/python3.14/site-packages/statsmodels/tsa/regime_switching/markov_switching.py:1292: EstimationWarning: Invalid regime transition probabilities estimated in EM iteration; probabilities have been re-scaled to continue estimation.\n  warnings.warn('Invalid regime transition probabilities'\n/Users/simon-pierreboucher/dev/vquant-next/.venv/lib/python3.14/site-packages/statsmodels/base/model.py:607: ConvergenceWarning: Maximum Likelihood optimization failed to converge. Check mle_retvals\n  warnings.warn(\"Maximum Likelihood optimization failed to \"\n/Users/simon-pierreboucher/dev/vquant-next/.venv/lib/python3.14/site-packages/statsmodels/tsa/regime_switching/markov_switching.py:1292: EstimationWarning: Invalid regime transition probabilities estimated in EM iteration; probabilities have been re-scaled to continue estimation.\n  warnings.warn('Invalid regime transition probabilities'\n/Users/simon-pierreboucher/dev/vquant-next/.venv/lib/python3.14/site-packages/statsmodels/tsa/regime_switching/markov_switching.py:1292: EstimationWarning: Invalid regime transition probabilities estimated in EM iteration; probabilities have been re-scaled to continue estimation.\n  warnings.warn('Invalid regime transition probabilities'\n/Users/simon-pierreboucher/dev/vquant-next/.venv/lib/python3.14/site-packages/statsmodels/tsa/regime_switching/markov_switching.py:1292: EstimationWarning: Invalid regime transition probabilities estimated in EM iteration; probabilities have been re-scaled to continue estimation.\n  warnings.warn('Invalid regime transition probabilities'\n/Users/simon-pierreboucher/dev/vquant-next/.venv/lib/python3.14/site-packages/statsmodels/tsa/regime_switching/markov_switching.py:1292: EstimationWarning: Invalid regime transition probabilities estimated in EM iteration; probabilities have been re-scaled to continue estimation.\n  warnings.warn('Invalid regime transition probabilities'\n/Users/simon-pierreboucher/dev/vquant-next/.venv/lib/python3.14/site-packages/statsmodels/tsa/regime_switching/markov_switching.py:1292: EstimationWarning: Invalid regime transition probabilities estimated in EM iteration; probabilities have been re-scaled to continue estimation.\n  warnings.warn('Invalid regime transition probabilities'\n/Users/simon-pierreboucher/dev/vquant-next/.venv/lib/python3.14/site-packages/statsmodels/tsa/regime_switching/markov_switching.py:1292: EstimationWarning: Invalid regime transition probabilities estimated in EM iteration; probabilities have been re-scaled to continue estimation.\n  warnings.warn('Invalid regime transition probabilities'\n/Users/simon-pierreboucher/dev/vquant-next/.venv/lib/python3.14/site-packages/statsmodels/tsa/regime_switching/markov_switching.py:1292: EstimationWarning: Invalid regime transition probabilities estimated in EM iteration; probabilities have been re-scaled to continue estimation.\n  warnings.warn('Invalid regime transition probabilities'\n/Users/simon-pierreboucher/dev/vquant-next/.venv/lib/python3.14/site-packages/statsmodels/tsa/regime_switching/markov_switching.py:1292: EstimationWarning: Invalid regime transition probabilities estimated in EM iteration; probabilities have been re-scaled to continue estimation.\n  warnings.warn('Invalid regime transition probabilities'\n/Users/simon-pierreboucher/dev/vquant-next/.venv/lib/python3.14/site-packages/statsmodels/tsa/regime_switching/markov_switching.py:1292: EstimationWarning: Invalid regime transition probabilities estimated in EM iteration; probabilities have been re-scaled to continue estimation.\n  warnings.warn('Invalid regime transition probabilities'\n/Users/simon-pierreboucher/dev/vquant-next/.venv/lib/python3.14/site-packages/statsmodels/tsa/regime_switching/markov_switching.py:1292: EstimationWarning: Invalid regime transition probabilities estimated in EM iteration; probabilities have been re-scaled to continue estimation.\n  warnings.warn('Invalid regime transition probabilities'\n/Users/simon-pierreboucher/dev/vquant-next/.venv/lib/python3.14/site-packages/statsmodels/base/model.py:607: ConvergenceWarning: Maximum Likelihood optimization failed to converge. Check mle_retvals\n  warnings.warn(\"Maximum Likelihood optimization failed to \"\n/Users/simon-pierreboucher/dev/vquant-next/.venv/lib/python3.14/site-packages/statsmodels/base/model.py:607: ConvergenceWarning: Maximum Likelihood optimization failed to converge. Check mle_retvals\n  warnings.warn(\"Maximum Likelihood optimization failed to \"\n/Users/simon-pierreboucher/dev/vquant-next/.venv/lib/python3.14/site-packages/statsmodels/base/model.py:607: ConvergenceWarning: Maximum Likelihood optimization failed to converge. Check mle_retvals\n  warnings.warn(\"Maximum Likelihood optimization failed to \"\n/Users/simon-pierreboucher/dev/vquant-next/.venv/lib/python3.14/site-packages/statsmodels/base/model.py:607: ConvergenceWarning: Maximum Likelihood optimization failed to converge. Check mle_retvals\n  warnings.warn(\"Maximum Likelihood optimization failed to \""}],"all_finished":true}},{"tool":"execute_custom_python_analysis","result":{"success":true,"output":"expanding n by h [119, 116, 112]\nrolling60 n by h [118, 115, 112]\nrolling40 n by h [117, 113, 111]\n      expanding               rolling60               rolling40              \n              1      4      8         1      4      8         1      4      8\nAR        1.000  1.000  1.000     1.000  1.000  1.000     1.000  1.000  1.000\nRW4       0.998  1.010  1.092     1.009  1.078  1.025     0.994  1.043  0.914\nIMA11     0.989  0.999  1.068     0.962  0.990  0.981     0.956  0.964  0.881\nUCSV      0.950  0.926  1.044     0.961  0.988  0.980     0.948  0.953  0.874\nMS_AR     1.045  1.105  1.104     1.017  1.083  1.038     1.020  1.011  0.953\nPC_u      1.052  1.028  1.156     1.050  1.118  1.390     1.042  1.125  1.228\nPC_ut     1.095  1.313  1.634     1.132  1.498  1.670     1.131  1.554  1.510\nVAR3      1.082  1.260  1.520     1.145  1.391  1.550     1.144  1.410  1.472\nRidge     1.034  1.259  1.468     1.001  1.392  1.556     1.029  1.378  1.479\nComb      0.980  1.022  1.120     1.002  1.091  1.118     0.987  1.068  1.024\n\n[Notebook session — variables in memory: df: DataFrame 203x13, calc: Series 203, load(), d: DataFrame 202x16, kpss(), zivot_andrews(), DFGLS(), PhillipsPerron(), rows: dict[6], v: VARResultsWrapper, x: DataFrame 112x15, name: str, sub: dict[4], adf: tuple[6], k: int, dg: float64, pp: float64, st: DataFrame 15x8, za: tuple[5], code: str, s: str, ms: str, uc: str, MarkovRegression(), y: ndarray (202,), X: ndarray (176, 5), mod: MarkovRegression, t: DataFrame 3x9, r: NoneType, CostLinear(), n: int, idx: PeriodIndex, MIN: int, bic_select(), res: dict[4], sig: ndarray (202, 1), o: list[5], kbest: tuple[4], sig_: ndarray (202, 1), pen: float64]","result":null,"figures":[],"files":[]}},{"tool":"execute_custom_python_analysis","result":{"success":true,"output":"sample  h    n                                                              MCS_set excluded_first\n0     all  1  119  AR, Comb, IMA11, MS_AR, MS_AR3, PC_u, PC_ut, RW4, Ridge, UCSV, VAR3               \n1     all  4  116  AR, Comb, IMA11, MS_AR, MS_AR3, PC_u, PC_ut, RW4, Ridge, UCSV, VAR3               \n2     all  8  112  AR, Comb, IMA11, MS_AR, MS_AR3, PC_u, PC_ut, RW4, Ridge, UCSV, VAR3               \n3  ex-GFC  1  114  AR, Comb, IMA11, MS_AR, MS_AR3, PC_u, PC_ut, RW4, Ridge, UCSV, VAR3               \n4  ex-GFC  4  111  AR, Comb, IMA11, MS_AR, MS_AR3, PC_u, PC_ut, RW4, Ridge, UCSV, VAR3               \n5  ex-GFC  8  107  AR, Comb, IMA11, MS_AR, MS_AR3, PC_u, PC_ut, RW4, Ridge, UCSV, VAR3               \nMAE rel AR\n            1      4      8\nAR     1.000  1.000  1.000\nRW4    1.036  1.052  0.989\nIMA11  1.028  0.987  0.956\nUCSV   0.967  0.950  0.938\nMS_AR  1.063  1.121  1.131\nPC_u   1.061  1.075  1.138\nPC_ut  1.089  1.264  1.430\nVAR3   1.076  1.239  1.323\nRidge  1.022  1.186  1.227\nComb   0.981  1.025  1.030\nOOS from 1985Q1, rel AR\n               1      4      8\nAR        1.000  1.000  1.000\nComb      0.974  0.951  0.902\nIMA11     0.991  0.965  0.902\nMS_AR     1.051  1.108  1.114\nMS_AR3    1.015  0.977  0.874\nMS_PC     1.123    NaN    NaN\nPC_gap1   1.015  1.017  1.028\nPC_u      1.070  0.986  1.029\nPC_ut     1.080  1.042  1.161\nRW1       1.186  1.344  1.427\nRW4       1.013  0.963  0.878\nRidge     1.029  1.001  1.141\nUCSV      0.929  0.843  0.821\nUCSV_g01  0.948  0.891  0.846\nVAR3      1.049  1.039  1.083\n\n[Notebook session — variables in memory: df: DataFrame 203x13, calc: Series 203, load(), d: DataFrame 202x16, kpss(), zivot_andrews(), DFGLS(), PhillipsPerron(), rows: dict[6], v: VARResultsWrapper, x: DataFrame 107x16, name: str, sub: dict[4], adf: tuple[6], k: int, dg: float64, pp: float64, st: DataFrame 15x8, za: tuple[5], code: str, s: str, ms: str, uc: str, MarkovRegression(), y: ndarray (202,), X: ndarray (176, 5), mod: MarkovRegression, t: DataFrame 3x9, r: NoneType, CostLinear(), n: int, idx: PeriodIndex, MIN: int, bic_select(), res: dict[4], sig: ndarray (202, 1), o: list[5], kbest: tuple[4], sig_: ndarray (202, 1), pen: float64]","result":null,"figures":[],"files":[]}},{"tool":"execute_custom_python_analysis","result":{"success":true,"output":"h   model   n    rel   DM_p\n0   1    UCSV  99  0.929  0.127\n1   1   IMA11  99  0.991  0.788\n2   1     RW4  99  1.013  0.583\n3   1  MS_AR3  99  1.015  0.390\n4   1    Comb  99  0.974  0.187\n5   1    PC_u  99  1.070  0.340\n6   1    VAR3  99  1.049  0.419\n7   4    UCSV  96  0.843  0.071\n8   4   IMA11  96  0.965  0.176\n9   4     RW4  96  0.963  0.412\n10  4  MS_AR3  96  0.977  0.465\n11  4    Comb  96  0.951  0.031\n12  4    PC_u  96  0.986  0.724\n13  4    VAR3  96  1.039  0.397\n14  8    UCSV  92  0.821  0.107\n15  8   IMA11  92  0.902  0.361\n16  8     RW4  92  0.878  0.295\n17  8  MS_AR3  92  0.874  0.020\n18  8    Comb  92  0.902  0.030\n19  8    PC_u  92  1.029  0.446\n20  8    VAR3  92  1.083  0.434\n1985+ 1 99 MCS: ['AR', 'Comb', 'IMA11', 'MS_AR', 'MS_AR3', 'PC_u', 'PC_ut', 'RW4', 'Ridge', 'UCSV', 'VAR3']\n1985+ 4 96 MCS: ['AR', 'Comb', 'IMA11', 'MS_AR', 'MS_AR3', 'PC_u', 'RW4', 'Ridge', 'UCSV', 'VAR3']\n1985+ 8 92 MCS: ['Comb', 'IMA11', 'MS_AR3', 'RW4', 'Ridge', 'UCSV']\n1985+ ex-GFC 1 94 MCS: ['AR', 'Comb', 'IMA11', 'MS_AR', 'MS_AR3', 'PC_u', 'PC_ut', 'RW4', 'Ridge', 'UCSV', 'VAR3']\n1985+ ex-GFC 4 91 MCS: ['AR', 'Comb', 'IMA11', 'MS_AR', 'MS_AR3', 'PC_u', 'PC_ut', 'RW4', 'Ridge', 'UCSV', 'VAR3']\n1985+ ex-GFC 8 87 MCS: ['Comb', 'IMA11', 'MS_AR3', 'RW4', 'Ridge', 'UCSV']\n\n[Notebook session — variables in memory: df: DataFrame 203x13, calc: Series 203, load(), d: DataFrame 202x16, kpss(), zivot_andrews(), DFGLS(), PhillipsPerron(), rows: list[21], v: VARResultsWrapper, x: DataFrame 87x16, name: str, sub: dict[4], adf: tuple[6], k: int, dg: float64, pp: float64, st: DataFrame 15x8, za: tuple[5], code: str, s: str, ms: str, uc: str, MarkovRegression(), y: ndarray (202,), X: ndarray (176, 5), mod: MarkovRegression, t: DataFrame 3x9, r: NoneType, CostLinear(), n: int, idx: PeriodIndex, MIN: int, bic_select(), res: dict[4], sig: ndarray (202, 1), o: list[5], kbest: tuple[4], sig_: ndarray (202, 1), pen: float64]","result":null,"figures":[],"files":[]}},{"tool":"execute_custom_python_analysis","result":{"success":true,"output":"PC [28.1 33.5]\nTR [48.  58.9]\n         est    se  persist\nend                        \n1969Q3 -0.99  0.29     0.29\n1974Q3  0.02  0.28     0.99\n1979Q3 -0.29  0.24     0.81\n1984Q3 -1.33  0.44     0.26\n1989Q3 -0.11  0.22     0.68\n1994Q3 -1.36  0.43    -0.24\n2009Q3 -0.50  0.48    -0.14\n         est    se  persist\nend                        \n1969Q4  0.84  0.27     0.65\n1974Q4  0.53  0.05     0.42\n1979Q4  0.68  0.18     0.60\n1984Q4  0.66  0.54     0.81\n1989Q4  0.72  0.23     0.53\n1994Q4 -0.57  0.54     0.86\n2009Q3  0.64  1.85     0.92\n\n[Notebook session — variables in memory: df: DataFrame 203x13, calc: Series 203, load(), d: DataFrame 202x16, kpss(), zivot_andrews(), DFGLS(), PhillipsPerron(), rows: list[21], v: VARResultsWrapper, x: DataFrame 87x16, name: str, sub: dict[4], adf: tuple[6], k: int, dg: float64, pp: float64, st: DataFrame 15x8, za: tuple[5], code: str, s: str, ms: str, uc: str, MarkovRegression(), y: ndarray (202,), X: ndarray (199, 4), mod: MarkovRegression, t: DataFrame 3x9, r: NoneType, CostLinear(), n: int, idx: PeriodIndex, MIN: int, bic_select(), res: dict[4], sig: ndarray (202, 1), o: list[5], kbest: tuple[4], sig_: ndarray (202, 1), pen: float64]","result":null,"figures":[],"files":[]}},{"tool":"execute_custom_python_analysis","result":{"success":true,"output":"\n\n---\nFIGURES PRODUCED BY THIS CALL (1) — ready to place in the report:\n\nFigure 1: ![Figure 1 : séries et ruptures détectées](/figures/fig-1795213288447-4aoc1eost-0.png)\n\n---\nPlace every figure you keep in the section it supports, using its markdown exactly as written above (same URL; the text in brackets is the caption shown under the image, so you may rewrite it as a short descriptive title, without the word \"Figure\" or a number: figures are numbered automatically in reading order). Announce the figure in the sentence before it, and follow it with a paragraph that reads it: what it shows, the number that matters, what it implies. A figure that does not serve the argument is left out rather than dropped at the end.\n\n[Figure check — VQuant visualization standard]\nFigure 1 — repaired automatically: contrast.\n[Vector copies (SVG, PDF) and the PNG of each figure are saved under artifacts/figures/ in the workspace.]\n\n[Notebook session — variables in memory: df: DataFrame 203x13, calc: Series 203, load(), d: DataFrame 202x16, kpss(), zivot_andrews(), DFGLS(), PhillipsPerron(), rows: list[21], v: str, x: DataFrame 87x16, name: str, sub: dict[4], adf: tuple[6], k: int, dg: float64, pp: float64, st: DataFrame 15x8, za: tuple[5], code: str, s: str, ms: str, uc: str, MarkovRegression(), y: ndarray (202,), X: ndarray (199, 4), mod: MarkovRegression, t: DataFrame 3x9, r: NoneType, CostLinear(), n: int, idx: PeriodIndex, MIN: int, bic_select(), res: dict[4], sig: ndarray (202,), o: list[5], kbest: tuple[4], sig_: ndarray (202, 1), pen: float64]","result":null,"files":[]}},{"tool":"execute_custom_python_analysis","result":{"success":true,"output":"\n\n---\nFIGURES PRODUCED BY THIS CALL (1) — ready to place in the report:\n\nFigure 2: ![Figure 1 corrigée](/figures/fig-1795213300346-hevonmefj-0.png)\n\n---\nPlace every figure you keep in the section it supports, using its markdown exactly as written above (same URL; the text in brackets is the caption shown under the image, so you may rewrite it as a short descriptive title, without the word \"Figure\" or a number: figures are numbered automatically in reading order). Announce the figure in the sentence before it, and follow it with a paragraph that reads it: what it shows, the number that matters, what it implies. A figure that does not serve the argument is left out rather than dropped at the end.\n\n[Figure check — VQuant visualization standard]\nFigure 1 — repaired automatically: contrast.\n[Vector copies (SVG, PDF) and the PNG of each figure are saved under artifacts/figures/ in the workspace.]\n\n[Notebook session — variables in memory: df: DataFrame 203x13, calc: Series 203, load(), d: DataFrame 202x16, kpss(), zivot_andrews(), DFGLS(), PhillipsPerron(), rows: list[21], v: str, x: DataFrame 87x16, name: str, sub: dict[4], adf: tuple[6], k: int, dg: float64, pp: float64, st: DataFrame 15x8, za: tuple[5], code: str, s: str, ms: str, uc: str, MarkovRegression(), y: ndarray (202,), X: ndarray (199, 4), mod: MarkovRegression, t: DataFrame 3x9, r: NoneType, CostLinear(), n: int, idx: PeriodIndex, MIN: int, bic_select(), res: dict[4], sig: ndarray (202,), o: list[5], kbest: tuple[4], sig_: ndarray (202, 1), pen: float64]","result":null,"files":[]}},{"tool":"execute_custom_python_analysis","result":{"success":true,"output":"\n\n---\nFIGURES PRODUCED BY THIS CALL (1) — ready to place in the report:\n\nFigure 3: ![Figure 2 : sup-Wald avec seuils bootstrap](/figures/fig-1795213300745-39k6qboz5-0.png)\n\n---\nPlace every figure you keep in the section it supports, using its markdown exactly as written above (same URL; the text in brackets is the caption shown under the image, so you may rewrite it as a short descriptive title, without the word \"Figure\" or a number: figures are numbered automatically in reading order). Announce the figure in the sentence before it, and follow it with a paragraph that reads it: what it shows, the number that matters, what it implies. A figure that does not serve the argument is left out rather than dropped at the end.\n\n[Figure check — VQuant visualization standard]\nFigure 1 — repaired automatically: contrast.\n[Vector copies (SVG, PDF) and the PNG of each figure are saved under artifacts/figures/ in the workspace.]\n\n[Notebook session — variables in memory: df: DataFrame 203x13, calc: Series 203, load(), d: DataFrame 202x16, kpss(), zivot_andrews(), DFGLS(), PhillipsPerron(), rows: list[21], v: str, x: DataFrame 87x16, name: str, sub: dict[4], adf: tuple[6], k: int, dg: float64, pp: float64, st: DataFrame 15x8, za: tuple[5], code: str, s: str, ms: str, uc: str, MarkovRegression(), y: ndarray (202,), X: ndarray (199, 4), mod: MarkovRegression, t: DataFrame 3x9, r: NoneType, CostLinear(), n: int, idx: PeriodIndex, MIN: int, bic_select(), res: dict[4], sig: ndarray (202,), o: list[5], kbest: tuple[4], sig_: ndarray (202, 1), pen: float64]","result":null,"files":[]}},{"tool":"execute_custom_python_analysis","result":{"success":true,"output":"\n\n---\nFIGURES PRODUCED BY THIS CALL (1) — ready to place in the report:\n\nFigure 4: ![Figure 4 : coefficients par régime](/figures/fig-1795213314285-a4pt5949s-0.png)\n\n---\nPlace every figure you keep in the section it supports, using its markdown exactly as written above (same URL; the text in brackets is the caption shown under the image, so you may rewrite it as a short descriptive title, without the word \"Figure\" or a number: figures are numbered automatically in reading order). Announce the figure in the sentence before it, and follow it with a paragraph that reads it: what it shows, the number that matters, what it implies. A figure that does not serve the argument is left out rather than dropped at the end.\n\n[Figure check — VQuant visualization standard]\nFigure 1 (“Pente de Phillips et réaction du taux court : des régimes bi”) — to fix in your code: Text elements overlap (for example “Règle de Taylor : coefficient ” and “Estimations OLS par régime (da”): move or shorten them.\nRedraw the figures that still have defects before using them in the report.\n[Vector copies (SVG, PDF) and the PNG of each figure are saved under artifacts/figures/ in the workspace.]\n\n[Notebook session — variables in memory: df: DataFrame 203x13, calc: Series 203, load(), d: DataFrame 202x16, kpss(), zivot_andrews(), DFGLS(), PhillipsPerron(), rows: list[21], v: str, x: DataFrame 87x16, name: str, sub: dict[4], adf: tuple[6], k: int, dg: float64, pp: float64, st: DataFrame 15x8, za: tuple[5], code: str, s: str, ms: str, uc: str, MarkovRegression(), y: ndarray (202,), X: ndarray (199, 4), mod: MarkovRegression, t: DataFrame 3x9, r: NoneType, CostLinear(), n: int, idx: PeriodIndex, MIN: int, bic_select(), res: dict[4], sig: ndarray (202,), o: list[5], kbest: tuple[4], sig_: ndarray (202, 1), pen: float64]","result":null,"files":[]}},{"tool":"execute_custom_python_analysis","result":{"success":true,"output":"\n\n---\nFIGURES PRODUCED BY THIS CALL (1) — ready to place in the report:\n\nFigure 5: ![Figure 5 : coefficients glissants](/figures/fig-1795213314712-gkdlkqd20-0.png)\n\n---\nPlace every figure you keep in the section it supports, using its markdown exactly as written above (same URL; the text in brackets is the caption shown under the image, so you may rewrite it as a short descriptive title, without the word \"Figure\" or a number: figures are numbered automatically in reading order). Announce the figure in the sentence before it, and follow it with a paragraph that reads it: what it shows, the number that matters, what it implies. A figure that does not serve the argument is left out rather than dropped at the end.\n\n[Figure check — VQuant visualization standard]\nFigure 1 — repaired automatically: contrast.\n[Vector copies (SVG, PDF) and the PNG of each figure are saved under artifacts/figures/ in the workspace.]\n\n[Notebook session — variables in memory: df: DataFrame 203x13, calc: Series 203, load(), d: DataFrame 202x16, kpss(), zivot_andrews(), DFGLS(), PhillipsPerron(), rows: list[21], v: str, x: DataFrame 87x16, name: str, sub: dict[4], adf: tuple[6], k: int, dg: float64, pp: float64, st: DataFrame 15x8, za: tuple[5], code: str, s: str, ms: str, uc: str, MarkovRegression(), y: ndarray (202,), X: ndarray (199, 4), mod: MarkovRegression, t: DataFrame 3x9, r: NoneType, CostLinear(), n: int, idx: PeriodIndex, MIN: int, bic_select(), res: dict[4], sig: ndarray (202,), o: list[5], kbest: tuple[4], sig_: ndarray (202, 1), pen: float64]","result":null,"files":[]}},{"tool":"execute_custom_python_analysis","result":{"success":true,"output":"\n\n---\nFIGURES PRODUCED BY THIS CALL (1) — ready to place in the report:\n\nFigure 6: ![Figure 4 corrigée](/figures/fig-1795213322098-9ha6rfkku-0.png)\n\n---\nPlace every figure you keep in the section it supports, using its markdown exactly as written above (same URL; the text in brackets is the caption shown under the image, so you may rewrite it as a short descriptive title, without the word \"Figure\" or a number: figures are numbered automatically in reading order). Announce the figure in the sentence before it, and follow it with a paragraph that reads it: what it shows, the number that matters, what it implies. A figure that does not serve the argument is left out rather than dropped at the end.\n\n[Figure check — VQuant visualization standard]\nFigure 1 — repaired automatically: contrast.\n[Vector copies (SVG, PDF) and the PNG of each figure are saved under artifacts/figures/ in the workspace.]\n\n[Notebook session — variables in memory: df: DataFrame 203x13, calc: Series 203, load(), d: DataFrame 202x16, kpss(), zivot_andrews(), DFGLS(), PhillipsPerron(), rows: list[21], v: str, x: DataFrame 87x16, name: str, sub: dict[4], adf: tuple[6], k: int, dg: float64, pp: float64, st: DataFrame 15x8, za: tuple[5], code: str, s: str, ms: str, uc: str, MarkovRegression(), y: ndarray (202,), X: ndarray (199, 4), mod: MarkovRegression, t: DataFrame 3x9, r: NoneType, CostLinear(), n: int, idx: PeriodIndex, MIN: int, bic_select(), res: dict[4], sig: ndarray (202,), o: list[5], kbest: tuple[4], sig_: ndarray (202, 1), pen: float64]","result":null,"files":[]}},{"tool":"execute_custom_python_analysis","result":{"success":true,"output":"regime means [5.98, 2.92] vol sd [3.69, 1.27]\nP(regime0) mean by period {'59-72': np.float64(0.04), '73-83': np.float64(0.93), '84-07': np.float64(0.22), '08-09': np.float64(0.92)}\ntau at 1980Q1, 1990Q1, 2000Q1, 2007Q4: [14.13, 4.75, 3.12, 3.29]\n\n---\nFIGURES PRODUCED BY THIS CALL (1) — ready to place in the report:\n\nFigure 7: ![Figure 6 : tendance UCSV et régimes markoviens](/figures/fig-1795213331748-blx6ysxga-0.png)\n\n---\nPlace every figure you keep in the section it supports, using its markdown exactly as written above (same URL; the text in brackets is the caption shown under the image, so you may rewrite it as a short descriptive title, without the word \"Figure\" or a number: figures are numbered automatically in reading order). Announce the figure in the sentence before it, and follow it with a paragraph that reads it: what it shows, the number that matters, what it implies. A figure that does not serve the argument is left out rather than dropped at the end.\n\n[Figure check — VQuant visualization standard]\nFigure 1 (“Deux lectures du changement de régime : tendance qui s'ancre”) — to fix in your code: Text elements overlap (for example “Tendance UCSV (filtrée, temps ” and “(haut) UCSV, filtre à particul”): move or shorten them.\nRedraw the figures that still have defects before using them in the report.\n[Vector copies (SVG, PDF) and the PNG of each figure are saved under artifacts/figures/ in the workspace.]\n\n[Notebook session — variables in memory: df: DataFrame 203x13, calc: Series 203, load(), d: DataFrame 202x16, kpss(), zivot_andrews(), DFGLS(), PhillipsPerron(), rows: list[21], v: str, x: DataFrame 87x16, name: str, sub: dict[4], adf: tuple[6], k: int, dg: float64, pp: float64, st: DataFrame 15x8, za: tuple[5], code: str, s: str, ms: str, uc: str, MarkovRegression(), y: ndarray (202,), X: ndarray (199, 4), mod: MarkovRegression, t: DataFrame 3x9, r: NoneType, CostLinear(), n: int, idx: PeriodIndex, MIN: int, bic_select(), res: dict[4], sig: ndarray (202,), o: list[5], kbest: tuple[4], sig_: ndarray (202, 1), pen: float64]","result":null,"files":[]}},{"tool":"execute_custom_python_analysis","result":{"success":true,"output":"\n\n---\nFIGURES PRODUCED BY THIS CALL (1) — ready to place in the report:\n\nFigure 8: ![Figure 6 corrigée](/figures/fig-1795213337724-l98nh8kv9-0.png)\n\n---\nPlace every figure you keep in the section it supports, using its markdown exactly as written above (same URL; the text in brackets is the caption shown under the image, so you may rewrite it as a short descriptive title, without the word \"Figure\" or a number: figures are numbered automatically in reading order). Announce the figure in the sentence before it, and follow it with a paragraph that reads it: what it shows, the number that matters, what it implies. A figure that does not serve the argument is left out rather than dropped at the end.\n\n[Figure check — VQuant visualization standard]\nFigure 1 — repaired automatically: contrast.\n[Vector copies (SVG, PDF) and the PNG of each figure are saved under artifacts/figures/ in the workspace.]\n\n[Notebook session — variables in memory: df: DataFrame 203x13, calc: Series 203, load(), d: DataFrame 202x16, kpss(), zivot_andrews(), DFGLS(), PhillipsPerron(), rows: list[21], v: str, x: DataFrame 87x16, name: str, sub: dict[4], adf: tuple[6], k: int, dg: float64, pp: float64, st: DataFrame 15x8, za: tuple[5], code: str, s: str, ms: str, uc: str, MarkovRegression(), y: ndarray (202,), X: ndarray (199, 4), mod: MarkovRegression, t: DataFrame 3x9, r: NoneType, CostLinear(), n: int, idx: PeriodIndex, MIN: int, bic_select(), res: dict[4], sig: ndarray (202,), o: list[5], kbest: tuple[4], sig_: ndarray (202, 1), pen: float64]","result":null,"files":[]}},{"tool":"execute_custom_python_analysis","result":{"success":true,"output":"1      4      8\nUCSV     0.950  0.926  1.044\nIMA11    0.989  0.999  1.068\nRW4      0.998  1.010  1.092\nComb     0.980  1.022  1.120\nMS_AR3   1.018  1.012  1.067\nPC_gap1  1.012  1.010  1.064\nMS_AR    1.045  1.105  1.104\nPC_u     1.052  1.028  1.156\nVAR3     1.082  1.260  1.520\nRidge    1.034  1.259  1.468\nPC_ut    1.095  1.313  1.634\n\n---\nFIGURES PRODUCED BY THIS CALL (1) — ready to place in the report:\n\nFigure 9: ![Figure 7 : RMSFE relatif par modèle et horizon](/figures/fig-1795213353082-wzprr6cat-0.png)\n\n---\nPlace every figure you keep in the section it supports, using its markdown exactly as written above (same URL; the text in brackets is the caption shown under the image, so you may rewrite it as a short descriptive title, without the word \"Figure\" or a number: figures are numbered automatically in reading order). Announce the figure in the sentence before it, and follow it with a paragraph that reads it: what it shows, the number that matters, what it implies. A figure that does not serve the argument is left out rather than dropped at the end.\n\n[Figure check — VQuant visualization standard]\nFigure 1 — repaired automatically: contrast.\n[Vector copies (SVG, PDF) and the PNG of each figure are saved under artifacts/figures/ in the workspace.]\n\n[Notebook session — variables in memory: df: DataFrame 203x13, calc: Series 203, load(), d: DataFrame 202x16, kpss(), zivot_andrews(), DFGLS(), PhillipsPerron(), rows: list[21], v: str, x: DataFrame 87x16, name: str, sub: dict[4], adf: tuple[6], k: int, dg: float64, pp: float64, st: DataFrame 15x8, za: tuple[5], code: str, s: str, ms: str, uc: str, MarkovRegression(), y: ndarray (202,), X: ndarray (199, 4), mod: MarkovRegression, t: DataFrame 3x9, r: NoneType, CostLinear(), n: int, idx: PeriodIndex, MIN: int, bic_select(), res: dict[4], sig: ndarray (202,), o: list[5], kbest: tuple[4], sig_: ndarray (202, 1), pen: float64]","result":null,"files":[]}},{"tool":"execute_custom_python_analysis","result":{"success":true,"output":"UCSV  IMA11  PC_u   VAR3  MS_AR3\norigin                                  \n2008Q3  54.9    0.8 -22.0 -226.1    -9.0\nUCSV csfe at 1989Q4, 2007Q4: [-20.7, 22.0] VAR3 1989Q4 -217.8\n\n---\nFIGURES PRODUCED BY THIS CALL (1) — ready to place in the report:\n\nFigure 10: ![Figure 8 : gains cumulés d'erreur quadratique (h=4)](/figures/fig-1795213353410-dfhbcelch-0.png)\n\n---\nPlace every figure you keep in the section it supports, using its markdown exactly as written above (same URL; the text in brackets is the caption shown under the image, so you may rewrite it as a short descriptive title, without the word \"Figure\" or a number: figures are numbered automatically in reading order). Announce the figure in the sentence before it, and follow it with a paragraph that reads it: what it shows, the number that matters, what it implies. A figure that does not serve the argument is left out rather than dropped at the end.\n\n[Figure check — VQuant visualization standard]\nFigure 1 (“Le gain de UCSV vient des années 1990 ; VAR et Phillips décr”) — to fix in your code: Text elements overlap (for example “crise 2008–09” and “Gain cumulé d'erreur quadratiq”): move or shorten them.\nRedraw the figures that still have defects before using them in the report.\n[Vector copies (SVG, PDF) and the PNG of each figure are saved under artifacts/figures/ in the workspace.]\n\n[Notebook session — variables in memory: df: DataFrame 203x13, calc: Series 203, load(), d: DataFrame 202x16, kpss(), zivot_andrews(), DFGLS(), PhillipsPerron(), rows: list[21], v: str, x: DataFrame 116x16, name: str, sub: dict[4], adf: tuple[6], k: int, dg: float64, pp: float64, st: DataFrame 15x8, za: tuple[5], code: str, s: str, ms: str, uc: str, MarkovRegression(), y: ndarray (202,), X: ndarray (199, 4), mod: MarkovRegression, t: DataFrame 3x9, r: NoneType, CostLinear(), n: int, idx: PeriodIndex, MIN: int, bic_select(), res: dict[4], sig: ndarray (202,), o: list[5], kbest: tuple[4], sig_: ndarray (202, 1), pen: float64]","result":null,"files":[]}},{"tool":"execute_custom_python_analysis","result":{"success":true,"output":"\n\n---\nFIGURES PRODUCED BY THIS CALL (1) — ready to place in the report:\n\nFigure 11: ![Figure 8 corrigée](/figures/fig-1795213363664-5zlb9f2a6-0.png)\n\n---\nPlace every figure you keep in the section it supports, using its markdown exactly as written above (same URL; the text in brackets is the caption shown under the image, so you may rewrite it as a short descriptive title, without the word \"Figure\" or a number: figures are numbered automatically in reading order). Announce the figure in the sentence before it, and follow it with a paragraph that reads it: what it shows, the number that matters, what it implies. A figure that does not serve the argument is left out rather than dropped at the end.\n\n[Figure check — VQuant visualization standard]\nFigure 1 — repaired automatically: contrast.\n[Vector copies (SVG, PDF) and the PNG of each figure are saved under artifacts/figures/ in the workspace.]\n\n[Notebook session — variables in memory: df: DataFrame 203x13, calc: Series 203, load(), d: DataFrame 202x16, kpss(), zivot_andrews(), DFGLS(), PhillipsPerron(), rows: list[21], v: str, x: DataFrame 116x16, name: str, sub: dict[4], adf: tuple[6], k: int, dg: float64, pp: float64, st: DataFrame 15x8, za: tuple[5], code: str, s: str, ms: str, uc: str, MarkovRegression(), y: ndarray (202,), X: ndarray (199, 4), mod: MarkovRegression, t: DataFrame 3x9, r: NoneType, CostLinear(), n: int, idx: PeriodIndex, MIN: int, bic_select(), res: dict[4], sig: ndarray (202,), o: list[5], kbest: tuple[4], sig_: ndarray (202, 1), pen: float64]","result":null,"files":[]}},{"tool":"execute_custom_python_analysis","result":{"success":true,"output":"True\n\n---\nFIGURES PRODUCED BY THIS CALL (1) — ready to place in the report:\n\nFigure 12: ![Figure 9 : carte de chaleur par sous-période](/figures/fig-1795213364173-fos0kmsg8-0.png)\n\n---\nPlace every figure you keep in the section it supports, using its markdown exactly as written above (same URL; the text in brackets is the caption shown under the image, so you may rewrite it as a short descriptive title, without the word \"Figure\" or a number: figures are numbered automatically in reading order). Announce the figure in the sentence before it, and follow it with a paragraph that reads it: what it shows, the number that matters, what it implies. A figure that does not serve the argument is left out rather than dropped at the end.\n\n[Figure check — VQuant visualization standard]\nFigure 1 — repaired automatically: contrast.\n[Vector copies (SVG, PDF) and the PNG of each figure are saved under artifacts/figures/ in the workspace.]\n\n[Notebook session — variables in memory: df: DataFrame 203x13, calc: Series 203, load(), d: DataFrame 202x16, kpss(), zivot_andrews(), DFGLS(), PhillipsPerron(), rows: list[21], v: str, x: DataFrame 116x16, name: str, sub: dict[4], adf: tuple[6], k: int, dg: float64, pp: float64, st: DataFrame 15x8, za: tuple[5], code: str, s: str, ms: str, uc: str, MarkovRegression(), y: ndarray (202,), X: ndarray (199, 4), mod: MarkovRegression, t: DataFrame 3x9, r: NoneType, CostLinear(), n: int, idx: PeriodIndex, MIN: int, bic_select(), res: dict[4], sig: ndarray (202,), o: list[5], kbest: tuple[4], sig_: ndarray (202, 1), pen: float64]","result":null,"files":[]}},{"tool":"execute_custom_python_analysis","result":{"success":true,"output":"\n\n---\nFIGURES PRODUCED BY THIS CALL (1) — ready to place in the report:\n\nFigure 13: ![Figure 9 titre corrigé](/figures/fig-1795213375061-fw6xmpnym-0.png)\n\n---\nPlace every figure you keep in the section it supports, using its markdown exactly as written above (same URL; the text in brackets is the caption shown under the image, so you may rewrite it as a short descriptive title, without the word \"Figure\" or a number: figures are numbered automatically in reading order). Announce the figure in the sentence before it, and follow it with a paragraph that reads it: what it shows, the number that matters, what it implies. A figure that does not serve the argument is left out rather than dropped at the end.\n\n[Figure check — VQuant visualization standard]\nFigure 1 — repaired automatically: contrast.\n[Vector copies (SVG, PDF) and the PNG of each figure are saved under artifacts/figures/ in the workspace.]\n\n[Notebook session — variables in memory: df: DataFrame 203x13, calc: Series 203, load(), d: DataFrame 202x16, kpss(), zivot_andrews(), DFGLS(), PhillipsPerron(), rows: list[21], v: str, x: DataFrame 116x16, name: str, sub: dict[4], adf: tuple[6], k: int, dg: float64, pp: float64, st: DataFrame 15x8, za: tuple[5], code: str, s: str, ms: str, uc: str, MarkovRegression(), y: ndarray (202,), X: ndarray (199, 4), mod: MarkovRegression, t: DataFrame 3x9, r: NoneType, CostLinear(), n: int, idx: PeriodIndex, MIN: int, bic_select(), res: dict[4], sig: ndarray (202,), o: list[5], kbest: tuple[4], sig_: ndarray (202, 1), pen: float64]","result":null,"files":[]}},{"tool":"execute_custom_python_analysis","result":{"success":true,"output":"\n\n---\nFIGURES PRODUCED BY THIS CALL (1) — ready to place in the report:\n\nFigure 14: ![Figure 10 : diagnostics des résidus (ACF)](/figures/fig-1795213375401-hoof4mdm0-0.png)\n\n---\nPlace every figure you keep in the section it supports, using its markdown exactly as written above (same URL; the text in brackets is the caption shown under the image, so you may rewrite it as a short descriptive title, without the word \"Figure\" or a number: figures are numbered automatically in reading order). Announce the figure in the sentence before it, and follow it with a paragraph that reads it: what it shows, the number that matters, what it implies. A figure that does not serve the argument is left out rather than dropped at the end.\n\n[Figure check — VQuant visualization standard]\nFigure 1 — repaired automatically: contrast.\n[Vector copies (SVG, PDF) and the PNG of each figure are saved under artifacts/figures/ in the workspace.]\n\n[Notebook session — variables in memory: df: DataFrame 203x13, calc: Series 203, load(), d: DataFrame 202x16, kpss(), zivot_andrews(), DFGLS(), PhillipsPerron(), rows: list[21], v: str, x: DataFrame 116x16, name: str, sub: dict[4], adf: tuple[6], k: int, dg: float64, pp: float64, st: DataFrame 15x8, za: tuple[5], code: str, s: str, ms: str, uc: str, MarkovRegression(), y: ndarray (202,), X: ndarray (199, 4), mod: MarkovRegression, t: DataFrame 3x9, r: NoneType, CostLinear(), n: int, idx: PeriodIndex, MIN: int, bic_select(), res: dict[4], sig: ndarray (202,), o: list[5], kbest: tuple[4], sig_: ndarray (202, 1), pen: float64]","result":null,"files":[]}},{"tool":"execute_custom_python_analysis","result":{"success":true,"output":"AR OOS mean error -0.343 p 0.049 LB4 p 0.014\nUCSV OOS mean error -0.253 p 0.183 LB4 p 0.105\nstd of 1-step AR err pre/post 1990: 2.76 1.6\n\n[Notebook session — variables in memory: df: DataFrame 203x13, calc: Series 203, load(), d: DataFrame 202x16, kpss(), zivot_andrews(), DFGLS(), PhillipsPerron(), rows: list[21], v: str, x: DataFrame 116x16, name: str, sub: dict[4], adf: tuple[6], k: int, dg: float64, pp: float64, st: DataFrame 15x8, za: tuple[5], code: str, s: str, ms: str, uc: str, MarkovRegression(), y: ndarray (202,), X: ndarray (199, 4), mod: MarkovRegression, t: DataFrame 3x9, r: NoneType, CostLinear(), n: int, idx: PeriodIndex, MIN: int, bic_select(), res: dict[4], sig: ndarray (202,), o: list[5], kbest: tuple[4], sig_: ndarray (202, 1), pen: float64]","result":null,"figures":[],"files":[]}},{"tool":"log_experiment","result":{"experiment_id":"4339e692-e113-46b1-98f0-ec1c8078075d","number":59,"status":"ok","recorded":true}},{"tool":"log_experiment","result":{"experiment_id":"aea472a2-15e7-4a04-8b29-d05371fc388c","number":60,"status":"ok","recorded":true}},{"tool":"log_experiment","result":{"experiment_id":"82b13f27-3807-404d-bb88-ffa254ff9158","number":61,"status":"ok","recorded":true}},{"tool":"log_experiment","result":{"experiment_id":"148b6706-4f19-4a52-ad01-b74ff1a0c262","number":62,"status":"failed","recorded":true}},{"tool":"log_experiment","result":{"experiment_id":"c046a30e-3425-4b1f-9277-c749ddb0361b","number":63,"status":"abandoned","recorded":true}},{"tool":"research_memory","result":{"saved":"4fc70c6c","kind":"finding","title":"US macro 1959-2009: breaks and inflation forecasting (results)"}},{"tool":"code_library","result":{"saved":"fc_eval","import":"from vqlib.fc_eval import load_all, add_comb, dm_test, cw_test","exports":["load_all","add_comb","dm_test","cw_test"],"note":"Importable in new notebook sessions and runs; a notebook session already running must importlib.reload it."}},{"tool":"execute_custom_python_analysis","result":{"success":true,"output":"-1.56 1984Q2 0.16 2005Q3 share windows |t|>1.96: 0.27\n-32.11 2004Q2 13.77 2002Q2 share windows LR>1 : 0.14 share ci excl 1 below: 0.36\n-0.33 1.02\n1979Q4 2009Q2\n\n[Notebook session — variables in memory: df: DataFrame 203x13, calc: Series 203, load(), d: DataFrame 202x16, kpss(), zivot_andrews(), DFGLS(), PhillipsPerron(), rows: list[21], v: str, x: DataFrame 119x16, name: str, sub: dict[4], adf: tuple[6], k: int, dg: float64, pp: float64, st: DataFrame 15x8, za: tuple[5], code: str, s: str, ms: str, uc: str, MarkovRegression(), y: ndarray (202,), X: ndarray (199, 4), mod: MarkovRegression, t: DataFrame 3x9, r: NoneType, CostLinear(), n: int, idx: PeriodIndex, MIN: int, bic_select(), res: dict[4], sig: ndarray (202,), o: list[5], kbest: tuple[4], sig_: ndarray (202, 1), pen: float64]","result":null,"figures":[],"files":[]}},{"tool":"log_experiment","result":{"experiment_id":"0282c8be-39a8-4f89-b738-24a235be164e","number":65,"status":"ok","recorded":true}}],"sources":[],"searchQueries":[],"pythonCode":null,"monteCarloResults":null,"customPythonFigures":[{"id":"fig-1795213288447-4aoc1eost","figures":["iVBORw0KGgoAAAANSUhEUgAACMYAAAckCAYAAACtc5ofAAAA…(trimmed 401840 chars)"],"figureUrls":["/figures/fig-1795213288447-4aoc1eost-0.png"],"output":"\n\n---\nFIGURES PRODUCED BY THIS CALL (1) — ready to place in the report:\n\nFigure 1: ![Figure 1 : séries et ruptures détectées](/figures/fig-1795213288447-4aoc1eost-0.png)\n\n---\nPlace every figure you keep in the section it supports, using its markdown exactly as written above (same URL; the text in brackets is the caption shown under the image, so you may rewrite it as a short descriptive title, without the word \"Figure\" or a number: figures are numbered automatically in reading order). Announce the figure in the sentence before it, and follow it with a paragraph that reads it: what it shows, the number that matters, what it implies. A figure that does not serve the argument is left out rather than dropped at the end.","description":"Figure 1 : séries et ruptures détectées"},{"id":"fig-1795213300346-hevonmefj","figures":["iVBORw0KGgoAAAANSUhEUgAACOgAAAckCAYAAAD856l8AAAA…(trimmed 401100 chars)"],"figureUrls":["/figures/fig-1795213300346-hevonmefj-0.png"],"output":"\n\n---\nFIGURES PRODUCED BY THIS CALL (1) — ready to place in the report:\n\nFigure 2: ![Figure 1 corrigée](/figures/fig-1795213300346-hevonmefj-0.png)\n\n---\nPlace every figure you keep in the section it supports, using its markdown exactly as written above (same URL; the text in brackets is the caption shown under the image, so you may rewrite it as a short descriptive title, without the word \"Figure\" or a number: figures are numbered automatically in reading order). Announce the figure in the sentence before it, and follow it with a paragraph that reads it: what it shows, the number that matters, what it implies. A figure that does not serve the argument is left out rather than dropped at the end.","description":"Figure 1 corrigée"},{"id":"fig-1795213300745-39k6qboz5","figures":["iVBORw0KGgoAAAANSUhEUgAACC0AAAYvCAYAAACuwZfuAAAA…(trimmed 257332 chars)"],"figureUrls":["/figures/fig-1795213300745-39k6qboz5-0.png"],"output":"\n\n---\nFIGURES PRODUCED BY THIS CALL (1) — ready to place in the report:\n\nFigure 3: ![Figure 2 : sup-Wald avec seuils bootstrap](/figures/fig-1795213300745-39k6qboz5-0.png)\n\n---\nPlace every figure you keep in the section it supports, using its markdown exactly as written above (same URL; the text in brackets is the caption shown under the image, so you may rewrite it as a short descriptive title, without the word \"Figure\" or a number: figures are numbered automatically in reading order). Announce the figure in the sentence before it, and follow it with a paragraph that reads it: what it shows, the number that matters, what it implies. A figure that does not serve the argument is left out rather than dropped at the end.","description":"Figure 2 : sup-Wald avec seuils bootstrap"},{"id":"fig-1795213314285-a4pt5949s","figures":["iVBORw0KGgoAAAANSUhEUgAACSoAAATUCAYAAABlWaukAAAA…(trimmed 243176 chars)"],"figureUrls":["/figures/fig-1795213314285-a4pt5949s-0.png"],"output":"\n\n---\nFIGURES PRODUCED BY THIS CALL (1) — ready to place in the report:\n\nFigure 4: ![Figure 4 : coefficients par régime](/figures/fig-1795213314285-a4pt5949s-0.png)\n\n---\nPlace every figure you keep in the section it supports, using its markdown exactly as written above (same URL; the text in brackets is the caption shown under the image, so you may rewrite it as a short descriptive title, without the word \"Figure\" or a number: figures are numbered automatically in reading order). Announce the figure in the sentence before it, and follow it with a paragraph that reads it: what it shows, the number that matters, what it implies. A figure that does not serve the argument is left out rather than dropped at the end.","description":"Figure 4 : coefficients par régime"},{"id":"fig-1795213314712-gkdlkqd20","figures":["iVBORw0KGgoAAAANSUhEUgAACBYAAAaBCAYAAAAdxJ6/AAAA…(trimmed 359644 chars)"],"figureUrls":["/figures/fig-1795213314712-gkdlkqd20-0.png"],"output":"\n\n---\nFIGURES PRODUCED BY THIS CALL (1) — ready to place in the report:\n\nFigure 5: ![Figure 5 : coefficients glissants](/figures/fig-1795213314712-gkdlkqd20-0.png)\n\n---\nPlace every figure you keep in the section it supports, using its markdown exactly as written above (same URL; the text in brackets is the caption shown under the image, so you may rewrite it as a short descriptive title, without the word \"Figure\" or a number: figures are numbered automatically in reading order). Announce the figure in the sentence before it, and follow it with a paragraph that reads it: what it shows, the number that matters, what it implies. A figure that does not serve the argument is left out rather than dropped at the end.","description":"Figure 5 : coefficients glissants"},{"id":"fig-1795213322098-9ha6rfkku","figures":["iVBORw0KGgoAAAANSUhEUgAACHoAAAT9CAYAAAA9YKbSAAAA…(trimmed 217440 chars)"],"figureUrls":["/figures/fig-1795213322098-9ha6rfkku-0.png"],"output":"\n\n---\nFIGURES PRODUCED BY THIS CALL (1) — ready to place in the report:\n\nFigure 6: ![Figure 4 corrigée](/figures/fig-1795213322098-9ha6rfkku-0.png)\n\n---\nPlace every figure you keep in the section it supports, using its markdown exactly as written above (same URL; the text in brackets is the caption shown under the image, so you may rewrite it as a short descriptive title, without the word \"Figure\" or a number: figures are numbered automatically in reading order). Announce the figure in the sentence before it, and follow it with a paragraph that reads it: what it shows, the number that matters, what it implies. A figure that does not serve the argument is left out rather than dropped at the end.","description":"Figure 4 corrigée"},{"id":"fig-1795213331748-blx6ysxga","figures":["iVBORw0KGgoAAAANSUhEUgAACk0AAAZYCAYAAAAxbRcmAAAA…(trimmed 396424 chars)"],"figureUrls":["/figures/fig-1795213331748-blx6ysxga-0.png"],"output":"regime means [5.98, 2.92] vol sd [3.69, 1.27]\nP(regime0) mean by period {'59-72': np.float64(0.04), '73-83': np.float64(0.93), '84-07': np.float64(0.22), '08-09': np.float64(0.92)}\ntau at 1980Q1, 1990Q1, 2000Q1, 2007Q4: [14.13, 4.75, 3.12, 3.29]\n\n---\nFIGURES PRODUCED BY THIS CALL (1) — ready to place in the report:\n\nFigure 7: ![Figure 6 : tendance UCSV et régimes markoviens](/figures/fig-1795213331748-blx6ysxga-0.png)\n\n---\nPlace every figure you keep in the section it supports, using its markdown exactly as written above (same URL; the text in brackets is the caption shown under the image, so you may rewrite it as a short descriptive title, without the word \"Figure\" or a number: figures are numbered automatically in reading order). Announce the figure in the sentence before it, and follow it with a paragraph that reads it: what it shows, the number that matters, what it implies. A figure that does not serve the argument is left out rather than dropped at the end.","description":"Figure 6 : tendance UCSV et régimes markoviens"},{"id":"fig-1795213337724-l98nh8kv9","figures":["iVBORw0KGgoAAAANSUhEUgAACBUAAAZYCAYAAADDwmhbAAAA…(trimmed 372940 chars)"],"figureUrls":["/figures/fig-1795213337724-l98nh8kv9-0.png"],"output":"\n\n---\nFIGURES PRODUCED BY THIS CALL (1) — ready to place in the report:\n\nFigure 8: ![Figure 6 corrigée](/figures/fig-1795213337724-l98nh8kv9-0.png)\n\n---\nPlace every figure you keep in the section it supports, using its markdown exactly as written above (same URL; the text in brackets is the caption shown under the image, so you may rewrite it as a short descriptive title, without the word \"Figure\" or a number: figures are numbered automatically in reading order). Announce the figure in the sentence before it, and follow it with a paragraph that reads it: what it shows, the number that matters, what it implies. A figure that does not serve the argument is left out rather than dropped at the end.","description":"Figure 6 corrigée"},{"id":"fig-1795213353082-wzprr6cat","figures":["iVBORw0KGgoAAAANSUhEUgAAB7EAAAV4CAYAAADGzSmdAAAA…(trimmed 224540 chars)"],"figureUrls":["/figures/fig-1795213353082-wzprr6cat-0.png"],"output":"1      4      8\nUCSV     0.950  0.926  1.044\nIMA11    0.989  0.999  1.068\nRW4      0.998  1.010  1.092\nComb     0.980  1.022  1.120\nMS_AR3   1.018  1.012  1.067\nPC_gap1  1.012  1.010  1.064\nMS_AR    1.045  1.105  1.104\nPC_u     1.052  1.028  1.156\nVAR3     1.082  1.260  1.520\nRidge    1.034  1.259  1.468\nPC_ut    1.095  1.313  1.634\n\n---\nFIGURES PRODUCED BY THIS CALL (1) — ready to place in the report:\n\nFigure 9: ![Figure 7 : RMSFE relatif par modèle et horizon](/figures/fig-1795213353082-wzprr6cat-0.png)\n\n---\nPlace every figure you keep in the section it supports, using its markdown exactly as written above (same URL; the text in brackets is the caption shown under the image, so you may rewrite it as a short descriptive title, without the word \"Figure\" or a number: figures are numbered automatically in reading order). Announce the figure in the sentence before it, and follow it with a paragraph that reads it: what it shows, the number that matters, what it implies. A figure that does not serve the argument is left out rather than dropped at the end.","description":"Figure 7 : RMSFE relatif par modèle et horizon"},{"id":"fig-1795213353410-dfhbcelch","figures":["iVBORw0KGgoAAAANSUhEUgAACGsAAAUmCAYAAAArgqtcAAAA…(trimmed 276828 chars)"],"figureUrls":["/figures/fig-1795213353410-dfhbcelch-0.png"],"output":"UCSV  IMA11  PC_u   VAR3  MS_AR3\norigin                                  \n2008Q3  54.9    0.8 -22.0 -226.1    -9.0\nUCSV csfe at 1989Q4, 2007Q4: [-20.7, 22.0] VAR3 1989Q4 -217.8\n\n---\nFIGURES PRODUCED BY THIS CALL (1) — ready to place in the report:\n\nFigure 10: ![Figure 8 : gains cumulés d'erreur quadratique (h=4)](/figures/fig-1795213353410-dfhbcelch-0.png)\n\n---\nPlace every figure you keep in the section it supports, using its markdown exactly as written above (same URL; the text in brackets is the caption shown under the image, so you may rewrite it as a short descriptive title, without the word \"Figure\" or a number: figures are numbered automatically in reading order). Announce the figure in the sentence before it, and follow it with a paragraph that reads it: what it shows, the number that matters, what it implies. A figure that does not serve the argument is left out rather than dropped at the end.","description":"Figure 8 : gains cumulés d'erreur quadratique (h=4)"},{"id":"fig-1795213363664-5zlb9f2a6","figures":["iVBORw0KGgoAAAANSUhEUgAACG0AAAUmCAYAAAAmnNsbAAAA…(trimmed 267328 chars)"],"figureUrls":["/figures/fig-1795213363664-5zlb9f2a6-0.png"],"output":"\n\n---\nFIGURES PRODUCED BY THIS CALL (1) — ready to place in the report:\n\nFigure 11: ![Figure 8 corrigée](/figures/fig-1795213363664-5zlb9f2a6-0.png)\n\n---\nPlace every figure you keep in the section it supports, using its markdown exactly as written above (same URL; the text in brackets is the caption shown under the image, so you may rewrite it as a short descriptive title, without the word \"Figure\" or a number: figures are numbered automatically in reading order). Announce the figure in the sentence before it, and follow it with a paragraph that reads it: what it shows, the number that matters, what it implies. A figure that does not serve the argument is left out rather than dropped at the end.","description":"Figure 8 corrigée"},{"id":"fig-1795213364173-fos0kmsg8","figures":["iVBORw0KGgoAAAANSUhEUgAACVcAAAVPCAYAAAC3OEi4AAAA…(trimmed 273812 chars)"],"figureUrls":["/figures/fig-1795213364173-fos0kmsg8-0.png"],"output":"True\n\n---\nFIGURES PRODUCED BY THIS CALL (1) — ready to place in the report:\n\nFigure 12: ![Figure 9 : carte de chaleur par sous-période](/figures/fig-1795213364173-fos0kmsg8-0.png)\n\n---\nPlace every figure you keep in the section it supports, using its markdown exactly as written above (same URL; the text in brackets is the caption shown under the image, so you may rewrite it as a short descriptive title, without the word \"Figure\" or a number: figures are numbered automatically in reading order). Announce the figure in the sentence before it, and follow it with a paragraph that reads it: what it shows, the number that matters, what it implies. A figure that does not serve the argument is left out rather than dropped at the end.","description":"Figure 9 : carte de chaleur par sous-période"},{"id":"fig-1795213375061-fw6xmpnym","figures":["iVBORw0KGgoAAAANSUhEUgAACJYAAAVPCAYAAADhy/LnAAAA…(trimmed 270156 chars)"],"figureUrls":["/figures/fig-1795213375061-fw6xmpnym-0.png"],"output":"\n\n---\nFIGURES PRODUCED BY THIS CALL (1) — ready to place in the report:\n\nFigure 13: ![Figure 9 titre corrigé](/figures/fig-1795213375061-fw6xmpnym-0.png)\n\n---\nPlace every figure you keep in the section it supports, using its markdown exactly as written above (same URL; the text in brackets is the caption shown under the image, so you may rewrite it as a short descriptive title, without the word \"Figure\" or a number: figures are numbered automatically in reading order). Announce the figure in the sentence before it, and follow it with a paragraph that reads it: what it shows, the number that matters, what it implies. A figure that does not serve the argument is left out rather than dropped at the end.","description":"Figure 9 titre corrigé"},{"id":"fig-1795213375401-hoof4mdm0","figures":["iVBORw0KGgoAAAANSUhEUgAACrsAAARaCAYAAAD2aFvaAAAA…(trimmed 174288 chars)"],"figureUrls":["/figures/fig-1795213375401-hoof4mdm0-0.png"],"output":"\n\n---\nFIGURES PRODUCED BY THIS CALL (1) — ready to place in the report:\n\nFigure 14: ![Figure 10 : diagnostics des résidus (ACF)](/figures/fig-1795213375401-hoof4mdm0-0.png)\n\n---\nPlace every figure you keep in the section it supports, using its markdown exactly as written above (same URL; the text in brackets is the caption shown under the image, so you may rewrite it as a short descriptive title, without the word \"Figure\" or a number: figures are numbered automatically in reading order). Announce the figure in the sentence before it, and follow it with a paragraph that reads it: what it shows, the number that matters, what it implies. A figure that does not serve the argument is left out rather than dropped at the end.","description":"Figure 10 : diagnostics des résidus (ACF)"}],"provenance":[{"tool":"inspect_dataset","provider":"python","symbols":[],"retrievedAt":1795212868298,"ok":true,"calls":1},{"tool":"research_run","provider":"python","symbols":[],"retrievedAt":1795213200592,"ok":true,"calls":10},{"tool":"execute_custom_python_analysis","provider":"python","symbols":[],"retrievedAt":1795213417023,"ok":true,"calls":46}],"model":"claude-sonnet-5-5","generationId":"0bc4b1ce-439f-451e-a893-f49987b53db6","usage":{"inputTokens":6620258,"outputTokens":78787,"costUsd":2.571177,"knownCostUsd":2.571177,"byModel":[{"model":"claude-sonnet-5-5","provider":"anthropic","role":"lead","calls":50,"inputTokens":6620258,"outputTokens":78787,"costUsd":2.571177}]},"research":{"workspaceId":"s-3d04acf2-235e-46e3-8c2d-bf26b0ba3c81","experiments":[{"id":"880c79ea-a091-43c4-bc0e-d38f5f81f624","seq":33,"title":"Stationarity battery (ADF, PP, DF-GLS, KPSS, Zivot-Andrews)","status":"ok","metrics":{"infl ADF p full":0.027,"infl ADF p 1959-83":0.456,"infl ADF p 1984-09":0,"unemp ADF p full":0.111,"tbill ADF p full":0.28,"ZA p infl":0.233},"error":null},{"id":"28b5d2e8-2c04-44db-b96e-e32e35a0cb42","seq":34,"title":"Break detection with ruptures (Dynp/PELT/Binseg/BottomUp) on levels and multivariate","status":"ok","metrics":{"infl PELT":"1973Q1, 1981Q4","infl Dynp BIC k=5":"1965Q4,1973Q1,1977Q4,1981Q4,1991Q1","tbill PELT":"1966Q1,1978Q3,1982Q3,1991Q1,2001Q3","unemp PELT":"1974Q4,1987Q2"},"error":null},{"id":"95294c79-17d5-4d34-b387-95e51f13fbea","seq":35,"title":"Regression breaks: sup-Wald with recursive wild bootstrap, Bai-Perron via CostLinear, Chow, OLS-CUSUM","status":"ok","metrics":{"PC supW all":44.4,"PC date":"1981Q4","PC p":0.004,"PC slope-only supW p":0.433,"TR supW all p":0.076,"TR date":"1980Q3","TR slope-only p":0.766,"OLS-CUSUM PC p":0.256,"OLS-CUSUM TR p":0.393},"error":null},{"id":"4339e692-e113-46b1-98f0-ec1c8078075d","seq":59,"title":"OOS forecasting horse race, expanding window (14 models, h=1,4,8)","status":"ok","metrics":{"RMSFE AR h1/h4/h8":"2.595/1.823/1.629","RMSFE UCSV":"2.466/1.688/1.701","rel UCSV":"0.950/0.926/1.044","rel PC_u":"1.052/1.028/1.156","rel VAR3":"1.082/1.260/1.520","rel MS_AR3":"1.018/1.012/1.067","DM p UCSV vs AR":"0.176/0.254/0.734"},"error":null},{"id":"aea472a2-15e7-4a04-8b29-d05371fc388c","seq":60,"title":"Robustness: rolling windows 60/40, MAE, start 1985, ex-GFC, subperiods, MCS, Phillips variants","status":"ok","metrics":{"post-1985 rel UCSV h1/4/8":"0.929/0.843/0.821","post-1985 DM p UCSV h4":0.071,"MCS 1985+ h8":"Comb, IMA11, MS_AR3, RW4, Ridge, UCSV (AR, PC, VAR excluded)","MCS full sample":"all 11 models retained at every horizon","MAE rel UCSV":"0.967/0.950/0.938","Phillips variants rel AR h1":"1.020/1.015/1.049"},"error":null},{"id":"82b13f27-3807-404d-bb88-ffa254ff9158","seq":61,"title":"Residual diagnostics (AR(3), Phillips ADL, Taylor, VAR(2), MS-AR(1))","status":"ok","metrics":{"AR3 LB4 p":0.951,"AR3 ARCH4 p":0,"PC LB4 p":0.0038,"Taylor kurtosis":17.6,"VAR whiteness p":0,"VAR Granger unemp->infl p":0.976,"VAR Granger tbill->infl p":0.0064,"MS-AR std. resid ARCH4 p":0.688,"MS-AR std. resid LB4 p":0},"error":null},{"id":"148b6706-4f19-4a52-ad01-b74ff1a0c262","seq":62,"title":"pip install ruptures via subprocess (failed attempt)","status":"failed","metrics":null,"error":null},{"id":"c046a30e-3425-4b1f-9277-c749ddb0361b","seq":63,"title":"Multivariate BIC break selection with ruptures (infl, unemp, tbill standardised)","status":"abandoned","metrics":null,"error":null},{"id":"0282c8be-39a8-4f89-b738-24a235be164e","seq":65,"title":"Rolling 40-quarter Phillips slope and Taylor long-run coefficient (HAC)","status":"ok","metrics":{"phillips_slope_min":-1.56,"phillips_slope_min_date":"1984Q2","phillips_slope_max":0.16,"phillips_slope_max_date":"2005Q3","share_windows_abs_t_gt_1.96":0.27,"taylor_share_windows_LR_gt_1":0.14,"taylor_share_CI_upper_below_1":0.36},"error":null}],"executions":{"total":56,"failed":4},"qualityGate":[{"id":"numbers","severity":"info","finding":"All 185 figures quoted in the text are present in the stored results (within rounding).","evidence":"2233 numbers in the stored results."},{"id":"sample","severity":"info","finding":"Sample size and period are stated for the statistical statements.","evidence":"Stated in the text."},{"id":"significance","severity":"warning","finding":"The text speaks of statistical significance, but no p-value is stored to support it.","evidence":"\"est significative\""},{"id":"figures","severity":"warning","finding":"3 figure(s) produced in this analysis still have defects the linter could not repair.","evidence":"Pente de Phillips et réaction du taux court : des régimes bien différents, mais des intervalles larges: Text elements overlap (for example “Règle de Taylor : coefficient ” and “Estimations OLS par régime (da”): move or shorten them. | Deux lectures du changement de régime : tendance qui s'ancre vers 2 % et régime volatil qui disparaît après 1986: Text elements overlap (for example “Tendance UCSV (filtrée, temps ” and “(haut) UCSV, filtre à particul”): move or shorten them. | Le gain de UCSV vient des années 1990 ; VAR et Phillips décrochent dans les années 1980: Text elements overlap (for example “crise 2008–09” and “Gain cumulé d'erreur quadratiq”): move or shorten them."},{"id":"failed_experiments","severity":"info","finding":"2 declared experiment(s) and 0 run(s) of this analysis did not succeed; the answer speaks of failures.","evidence":"#62 pip install ruptures via subprocess (failed attempt) [failed]: pip through subprocess failed with SSL certificate error; install_package(\"ruptures\") worked (v1.1.10). No effect on res | #63 Multivariate BIC break selection with ruptures (infl, unemp, tbill standardised) [abandoned]: BIC decreased monotonically up to k=6 (never reached a minimum), so no break count could be chosen; dropped in favour of"}]}},{"question":"Le serveur a redémarré depuis ton étude : ton notebook est vide, mais le workspace est intact. Reprends le projet sans tout recalculer : à partir des prévisions déjà enregistrées, ajoute le test de fluctuation de Giacomini-Rossi (stabilité dans le temps de la précision relative UCSV contre AR, h=4) que tu avais laissé de côté, en programmant toi-même la statistique et en prenant les valeurs critiques dans l article ou par simulation, puis donne une figure de qualité publication et une conclusion courte. Dis ce que tu as réutilisé du workspace.","answer":"\n\nReproduit à l'identique par un script autonome. Je consigne le tout.\n\n## Conclusion\n\nLe test de Giacomini-Rossi ne détecte aucune instabilité de la précision relative UCSV/AR à h=4. Il n'établit pas pour autant un avantage durable d'UCSV : cet avantage est presque entièrement dû aux cibles de 2005 à 2009.\n\nSur la fenêtre de 35 trimestres (μ = 0,30), la statistique maximale vaut |F|max = 2,39, pour un seuil de 3,07 à 5 % et 2,81 à 10 %. Sur la fenêtre de 46 trimestres, elle vaut 2,01 pour un seuil de 2,93. Aucune des 36 spécifications testées ne rejette l'hypothèse nulle.\n\n## Données et méthode\n\n- **Échantillon :** 116 prévisions à quatre trimestres, avec des origines de 1979T4 à 2008T3. Le rapport des RMSFE est de 0,926 et le test de Diebold-Mariano donne p = 0,254, comme dans l'étude initiale.\n- **Différence de perte :** ΔL = e²_AR − e²_UCSV. Une valeur positive signifie qu'UCSV est meilleur.\n- **Statistique :** F_t est la somme glissante de ΔL sur m trimestres, divisée par σ̂·√m. σ̂ est un écart-type de long terme de Bartlett (Newey-West). La statistique est codée à la main dans `code/gr_fluct.py`.\n- **Seuils :** simulés (60 000 réplications) à partir de la limite brownienne, avec le μ exact de chaque fenêtre. Les seuils suivent la table de l'article, mais mes simulations sont plus hautes de 0,05 à 0,09. Je cite la table de mémoire, sans avoir rouvert le PDF, et j'ai retenu mes valeurs simulées, plus prudentes.\n- **Spécifications testées :** m ∈ {29, 35, 46, 58}, troncature L ∈ {3, 4, 8}, perte quadratique ou absolue, UCSV avec γ = 0,1 ou 0,2. Les rejets sont comptés sur les 36 combinaisons retenues (4 fenêtres × 3 troncatures pour chacun des trois cas : perte quadratique γ = 0,2, perte absolue γ = 0,2, perte quadratique γ = 0,1).\n- **Choix de la troncature :** je comptais utiliser L = 4. Un contrôle de taille sur des pertes simulées en MA(3), avec 116 observations, montre qu'avec L = 3 ou 4 le test rejette à tort 8 à 15 % du temps pour un seuil nominal de 5 %. Avec L = 8 la taille est de 5 %. J'ai donc retenu L = 8. Le non-rejet vaut aussi pour L = 3 et 4.\n- **Puissance :** environ 80 % pour un décalage de moyenne de 0,5 écart-type de long terme au milieu de l'échantillon. Un non-rejet exclut donc les instabilités nettes, pas les petites.\n\n![UCSV contre AR à h=4 : gain cumulé en erreur quadratique (haut) et statistique de fluctuation de Giacomini-Rossi avec ses seuils à 5 et 10 % (bas), 1980–2009](/figures/fig-1795213857744-yrr6kz9wk-0.png)\n\nDans le panneau du haut, la courbe descend jusqu'à environ −30 vers 1983 : l'AR domine au début. Elle remonte ensuite, mais surtout à la fin. Par blocs de cinq ans, le gain cumulé est de −25 pour les cibles 1980-84, +4, +10, +7 et +1 pour les quatre blocs suivants, puis +59 pour 2005-09. La statistique F_t du panneau du bas reste dans la bande de non-rejet, avec un maximum au centre de fenêtre de mars 2005.\n\nQuatre trimestres portent l'essentiel du gain : 2009T2 (27), 2006T3 (14), 2008T4 (11) et 2009T1 (8,5). Si l'on exclut les cibles de 2005 à 2009, le rapport des RMSFE passe à 1,012 (Diebold-Mariano p = 0,80). Sur 2005-09 seul (16 observations), il est de 0,77.\n\nCela corrige une phrase de mon étude précédente, qui situait le gain d'UCSV dans les années 1990 : par blocs de cinq ans, ces années-là contribuent peu. Il reste que le test ne rejette pas la stabilité. Je lis le résultat ainsi : la précision relative est trop bruitée pour qu'on puisse affirmer qu'elle a changé, et trop peu documentée en dehors des chocs de 2005-09 pour qu'on puisse affirmer qu'UCSV bat l'AR. C'est mon interprétation.\n\n## Ce que j'ai réutilisé du workspace\n\n- **Prévisions enregistrées, sans rien recalculer :** `artifacts/forecasts/linear_expanding.csv` (AR) et `ucsv_g0.2.csv` / `ucsv_g0.1.csv` (UCSV).\n- **Code :** `code/evalfc.py` pour le test de Diebold-Mariano corrigé (HLN), utilisé pour retrouver p = 0,254.\n- **Mémoire de recherche :** les conventions de l'étude (UCSV de référence, h=4, fenêtre extensible) et le rapport relatif de 0,93.\n- **Nouveau :** `code/gr_fluct.py`, `code/gr_run.py` (script autonome qui reproduit les chiffres) et `artifacts/gr_results.csv`, `gr_critical_values.csv`, `gr_size_check.csv`, `gr_power_check.csv`. Le graphique est `artifacts/figures/gr_fluctuation_ucsv_vs_ar_h4.png`, au format vectoriel dans le même dossier. L'expérience n° 82 et la note de mémoire sont mises à jour.\n\n## Limites\n\n- L'échantillon est court (P = 116, avec des erreurs qui se chevauchent) et le test de taille est surdimensionné pour L = 3 ou 4.\n- Les seuils sont simulés, pas relus dans l'article.\n- Les prévisions s'arrêtent en 2009T3.\n\nSources : statsmodels macrodata (données trimestrielles américaines 1959-2009), prévisions enregistrées du workspace, calculs propres.\n\nFausse alerte : les « 60 000 réplications » sont le paramètre `reps=60000` que j'ai passé à `sim_cv` pour les seuils utilisés (cellule de calcul et `code/gr_run.py`). Ce n'est pas un résultat qui aurait dû être stocké, d'où l'absence dans les sorties enregistrées.","toolResults":[{"tool":"research_memory","result":{"notes":[{"id":"4fc70c6c","kind":"finding","title":"US macro 1959-2009: breaks and inflation forecasting (results)","body":"Data: data/us_macro_quarterly.csv (statsmodels macrodata), drop first row (infl=0 placeholder); infl = 400*dlog(cpi). code/common.py load(); code/fcsuite.py; fc_linear.py, fc_ms.py, fc_ms3.py, fc_ucsv.py write artifacts/forecasts/*.csv; evalfc.py has DM-HLN and Clark-West. Breaks: inflation mean shifts 1973Q1 and 1981Q4 (PELT, robust); Phillips ADL break 1981Q4 in intercept/persistence (sup-Wald wild bootstrap p=0.004) not in unemployment slope (p=0.43); Taylor rule break marginal (p=0.076), BP dates 1980Q4/1985Q1. OOS 1980Q1-2009Q3: AR(BIC) hard to beat; UCSV best h=1,4 (rel RMSFE 0.95/0.93, DM p 0.18/0.25); Phillips/VAR/Ridge worse; MS-AR no gain; MCS keeps all models in full sample. Post-1985 UCSV rel 0.93/0.84/0.82 (DM p 0.13/0.07/0.11). MS rolling fits unstable (fail often).","updated":"2026-11-20"}],"kinds":["dataset","model","decision","implementation","failure","hypothesis","finding","workflow"]}},{"tool":"workspace_files","result":{"workspace":"s-3d04acf2-235e-46e3-8c2d-bf26b0ba3c81","files":["artifacts/","artifacts/dm_post1985.csv  (1041 bytes)","artifacts/dm_vs_AR_expanding.csv  (3164 bytes)","artifacts/figures/","artifacts/figures/20261120-222127-01-trois-s-ries-des-moyennes-par-paliers-l-inflatio.pdf  (28066 bytes)","artifacts/figures/20261120-222127-01-trois-s-ries-des-moyennes-par-paliers-l-inflatio.png  (301380 bytes)","artifacts/figures/20261120-222127-01-trois-s-ries-des-moyennes-par-paliers-l-inflatio.svg  (114004 bytes)","artifacts/figures/20261120-222140-01-l-inflation-bascule-en-1973t1-et-1981t4-ch-mage-.pdf  (27490 bytes)","artifacts/figures/20261120-222140-01-l-inflation-bascule-en-1973t1-et-1981t4-ch-mage-.png  (300824 bytes)","artifacts/figures/20261120-222140-01-l-inflation-bascule-en-1973t1-et-1981t4-ch-mage-.svg  (113049 bytes)","artifacts/figures/20261120-222140-01-la-courbe-de-phillips-rompt-en-1981t4-la-r-gle-d.pdf  (24169 bytes)","artifacts/figures/20261120-222140-01-la-courbe-de-phillips-rompt-en-1981t4-la-r-gle-d.png  (192999 bytes)","artifacts/figures/20261120-222140-01-la-courbe-de-phillips-rompt-en-1981t4-la-r-gle-d.svg  (104559 bytes)","artifacts/figures/20261120-222154-01-estimations-sur-fen-tres-glissantes-de-40-trimes.pdf  (30265 bytes)","artifacts/figures/20261120-222154-01-estimations-sur-fen-tres-glissantes-de-40-trimes.png  (269732 bytes)","artifacts/figures/20261120-222154-01-estimations-sur-fen-tres-glissantes-de-40-trimes.svg  (120968 bytes)","artifacts/figures/20261120-222154-01-pente-de-phillips-et-r-action-du-taux-court-des-.pdf  (22901 bytes)","artifacts/figures/20261120-222154-01-pente-de-phillips-et-r-action-du-taux-court-des-.png  (182382 bytes)","artifacts/figures/20261120-222154-01-pente-de-phillips-et-r-action-du-taux-court-des-.svg  (118068 bytes)","artifacts/figures/20261120-222201-01-pente-de-phillips-et-r-action-du-taux-court-r-gi.pdf  (22241 bytes)","artifacts/figures/20261120-222201-01-pente-de-phillips-et-r-action-du-taux-court-r-gi.png  (163080 bytes)","artifacts/figures/20261120-222201-01-pente-de-phillips-et-r-action-du-taux-court-r-gi.svg  (116273 bytes)","artifacts/figures/20261120-222211-01-deux-lectures-du-changement-de-r-gime-tendance-q.pdf  (28426 bytes)","artifacts/figures/20261120-222211-01-deux-lectures-du-changement-de-r-gime-tendance-q.png  (297318 bytes)","artifacts/figures/20261120-222211-01-deux-lectures-du-changement-de-r-gime-tendance-q.svg  (123005 bytes)","artifacts/figures/20261120-222217-01-la-tendance-d-inflation-s-ancre-vers-3-et-le-r-g.pdf  (27675 bytes)","artifacts/figures/20261120-222217-01-la-tendance-d-inflation-s-ancre-vers-3-et-le-r-g.png  (279705 bytes)","artifacts/figures/20261120-222217-01-la-tendance-d-inflation-s-ancre-vers-3-et-le-r-g.svg  (117176 bytes)","artifacts/figures/20261120-222232-01-aucun-mod-le-conomique-ne-bat-l-ar-seuls-ucsv-et.pdf  (27680 bytes)","artifacts/figures/20261120-222232-01-aucun-mod-le-conomique-ne-bat-l-ar-seuls-ucsv-et.png  (168403 bytes)","artifacts/figures/20261120-222232-01-aucun-mod-le-conomique-ne-bat-l-ar-seuls-ucsv-et.svg  (121229 bytes)","artifacts/figures/20261120-222233-01-le-gain-de-ucsv-vient-des-ann-es-1990-var-et-phi.pdf  (28679 bytes)","artifacts/figures/20261120-222233-01-le-gain-de-ucsv-vient-des-ann-es-1990-var-et-phi.png  (207621 bytes)","artifacts/figures/20261120-222233-01-le-gain-de-ucsv-vient-des-ann-es-1990-var-et-phi.svg  (116540 bytes)","artifacts/figures/20261120-222243-01-le-classement-s-inverse-avant-1990-l-ar-domine-a.pdf  (36453 bytes)","artifacts/figures/20261120-222243-01-le-classement-s-inverse-avant-1990-l-ar-domine-a.png  (205357 bytes)","artifacts/figures/20261120-222243-01-le-classement-s-inverse-avant-1990-l-ar-domine-a.svg  (142546 bytes)","artifacts/figures/20261120-222243-01-le-gain-de-ucsv-vient-des-ann-es-1990-var-et-phi.pdf  (27050 bytes)","artifacts/figures/20261120-222243-01-le-gain-de-ucsv-vient-des-ann-es-1990-var-et-phi.png  (200495 bytes)","artifacts/figures/20261120-222243-01-le-gain-de-ucsv-vient-des-ann-es-1990-var-et-phi.svg  (115705 bytes)","artifacts/figures/20261120-222254-01-avant-1990-l-ar-est-dur-battre-ensuite-les-mod-l.pdf  (35678 bytes)","artifacts/figures/20261120-222254-01-avant-1990-l-ar-est-dur-battre-ensuite-les-mod-l.png  (202617 bytes)","artifacts/figures/20261120-222254-01-avant-1990-l-ar-est-dur-battre-ensuite-les-mod-l.svg  (142199 bytes)","artifacts/figures/20261120-222255-01-les-r-sidus-sont-blancs-pour-l-ar-3-mais-h-t-ros.pdf  (23558 bytes)","artifacts/figures/20261120-222255-01-les-r-sidus-sont-blancs-pour-l-ar-3-mais-h-t-ros.png  (130715 bytes)","artifacts/figures/20261120-222255-01-les-r-sidus-sont-blancs-pour-l-ar-3-mais-h-t-ros.svg  (114520 bytes)","artifacts/forecasts/","artifacts/forecasts/linear_expanding.csv  (127280 bytes)","artifacts/forecasts/linear_rolling40.csv  (127094 bytes)","artifacts/forecasts/linear_rolling60.csv  (127172 bytes)","artifacts/forecasts/ms3_expanding.csv  (15638 bytes)","artifacts/forecasts/ms3_rolling60.csv  (15370 bytes)","artifacts/forecasts/ms_expanding.csv  (19820 bytes)","artifacts/forecasts/ms_expanding_probs.csv  (5505 bytes)","artifacts/forecasts/ms_rolling40.csv  (19246 bytes)","artifacts/forecasts/ms_rolling60.csv  (19472 bytes)","artifacts/forecasts/pc_variants_expanding.csv  (28692 bytes)","artifacts/forecasts/ucsv_g0.1.csv  (14951 bytes)","artifacts/forecasts/ucsv_g0.1_tau.csv  (8897 bytes)","artifacts/forecasts/ucsv_g0.2.csv  (14967 bytes)","artifacts/forecasts/ucsv_g0.2_tau.csv  (8876 bytes)","artifacts/mcs_results.csv  (529 bytes)","artifacts/pc_regimes.csv  (219 bytes)","artifacts/residual_diagnostics.csv  (527 bytes)","artifacts/rmsfe_expanding.csv  (349 bytes)","artifacts/rmsfe_rel_expanding.csv  (261 bytes)","artifacts/robust_mae_rel.csv  (579 bytes)","artifacts/robust_start1985_rel.csv  (862 bytes)","artifacts/robust_window_rmsfe.csv  (1838 bytes)","artifacts/rolling_pc.csv  (10674 bytes)","artifacts/rolling_taylor.csv  (10313 bytes)","artifacts/stationarity.csv  (735 bytes)","artifacts/subsample_results.csv  (8433 bytes)","artifacts/taylor_regimes.csv  (240 bytes)","code/","code/common.py  (389 bytes)","code/evalfc.py  (1608 bytes)","code/fc_linear.py  (3973 bytes)","code/fc_ms.py  (3150 bytes)","code/fc_ms3.py  (2120 bytes)","code/fc_ucsv.py  (1323 bytes)","code/fcsuite.py  (1267 bytes)","data/","data/us_macro_quarterly.csv  (17424 bytes)","environment.lock.txt  (1535 bytes)","logs/","logs/cells/","logs/cells/0519105012f0bf0b.py  (235 bytes)","logs/cells/05e0b4ebc9be4a06.py  (2470 bytes)","logs/cells/0e56ad33d5c77039.py  (2167 bytes)","logs/cells/1c3cb6a27883ca38.py  (1517 bytes)","logs/cells/2022b583b9b7d1d7.py  (156 bytes)","logs/cells/21f914e1b8841466.py  (2213 bytes)","logs/cells/2234ca76ed293db4.py  (1665 bytes)","logs/cells/2448361473081301.py  (801 bytes)","logs/cells/25878954f157274a.py  (2329 bytes)","logs/cells/46f2a429205aece1.py  (4025 bytes)","logs/cells/4e341082f7303cc8.py  (1473 bytes)","logs/cells/4ebc8ee4e1c14edc.py  (830 bytes)","logs/cells/507fac47704dc75d.py  (1180 bytes)","logs/cells/51c66189f7656928.py  (594 bytes)","logs/cells/59d6eae530543edb.py  (1426 bytes)","logs/cells/5bd91f0e84404512.py  (1128 bytes)","logs/cells/5f8fd6b88c01396c.py  (1758 bytes)","logs/cells/66551dbd7a39604b.py  (1771 bytes)","logs/cells/6d03bac6ff8e1336.py  (1200 bytes)","logs/cells/7a3ed8a0fe92bea2.py  (495 bytes)","logs/cells/801ee2c9fa522624.py  (964 bytes)","logs/cells/832b65722c50ff55.py  (1240 bytes)","logs/cells/87b7a3d7e24e76a0.py  (476 bytes)","logs/cells/8d9c3f020872866c.py  (759 bytes)","logs/cells/9318c7d73701bd4f.py  (750 bytes)","logs/cells/9a2943f040748ecc.py  (1304 bytes)","logs/cells/a062c84a82242a0d.py  (1226 bytes)","logs/cells/a12580318c97fb04.py  (1162 bytes)","logs/cells/aa5ac2b7d296ccc0.py  (74 bytes)","logs/cells/b7c170daa114d86e.py  (824 bytes)","logs/cells/c0153bfbb0eac1c2.py  (591 bytes)","logs/cells/c0fb517c33ca2f5a.py  (846 bytes)","logs/cells/c21b362971a36581.py  (923 bytes)","logs/cells/c59e322ba4dd65e4.py  (994 bytes)","logs/cells/cc52d3ae860278f1.py  (684 bytes)","logs/cells/cd49d9614156e834.py  (392 bytes)","logs/cells/cfb8a08eeb7aaf70.py  (1323 bytes)","logs/cells/dbf3a997ec442aa0.py  (2343 bytes)","logs/cells/e8250ace1e808391.py  (863 bytes)","logs/cells/eb9b21b54083174e.py  (4573 bytes)","logs/cells/ec89c9873d6b0717.py  (1575 bytes)","logs/cells/f2b91175604d3d0d.py  (2246 bytes)","logs/cells/f45932d48cd510b5.py  (2614 bytes)","logs/cells/f65d4f5639b0e3aa.py  (174 bytes)","logs/cells/f7abfa9ef255f2cf.py  (1669 bytes)","logs/cells/fad5c8858602ff90.py  (1090 bytes)","logs/executions.jsonl  (18363 bytes)","runs/","runs/0b4b4569-3c89-4bdf-b7c9-ff5a75c1afff/","runs/0b4b4569-3c89-4bdf-b7c9-ff5a75c1afff/run.json  (393 bytes)","runs/0b4b4569-3c89-4bdf-b7c9-ff5a75c1afff/status.json  (97 bytes)","runs/0b4b4569-3c89-4bdf-b7c9-ff5a75c1afff/stderr.log  (0 bytes)","runs/0b4b4569-3c89-4bdf-b7c9-ff5a75c1afff/stdout.log  (85 bytes)","runs/64827983-99cb-4ddf-a37e-2f7184f388d4/","runs/64827983-99cb-4ddf-a37e-2f7184f388d4/run.json  (414 bytes)","runs/64827983-99cb-4ddf-a37e-2f7184f388d4/status.json  (96 bytes)","runs/64827983-99cb-4ddf-a37e-2f7184f388d4/stderr.log  (0 bytes)","runs/64827983-99cb-4ddf-a37e-2f7184f388d4/stdout.log  (382 bytes)","runs/8125167d-cec8-434f-b799-c5365fcf2a6c/","runs/8125167d-cec8-434f-b799-c5365fcf2a6c/run.json  (400 bytes)","runs/8125167d-cec8-434f-b799-c5365fcf2a6c/status.json  (96 bytes)","runs/8125167d-cec8-434f-b799-c5365fcf2a6c/stderr.log  (9492 bytes)","runs/8125167d-cec8-434f-b799-c5365fcf2a6c/stdout.log  (169 bytes)","runs/8f82b8f5-d691-4ba0-b8fa-cb325149fc11/","runs/8f82b8f5-d691-4ba0-b8fa-cb325149fc11/run.json  (400 bytes)","runs/8f82b8f5-d691-4ba0-b8fa-cb325149fc11/status.json  (97 bytes)","runs/8f82b8f5-d691-4ba0-b8fa-cb325149fc11/stderr.log  (53048 bytes)","runs/8f82b8f5-d691-4ba0-b8fa-cb325149fc11/stdout.log  (171 bytes)","runs/908d8db9-e156-49bd-b54f-a3b0eee8db27/","runs/908d8db9-e156-49bd-b54f-a3b0eee8db27/run.json  (401 bytes)","runs/908d8db9-e156-49bd-b54f-a3b0eee8db27/status.json  (95 bytes)","runs/908d8db9-e156-49bd-b54f-a3b0eee8db27/stderr.log  (0 bytes)","runs/908d8db9-e156-49bd-b54f-a3b0eee8db27/stdout.log  (382 bytes)","runs/a0e1b107-631d-489d-a128-c8f32694c432/","runs/a0e1b107-631d-489d-a128-c8f32694c432/run.json  (400 bytes)","runs/a0e1b107-631d-489d-a128-c8f32694c432/status.json  (96 bytes)","runs/a0e1b107-631d-489d-a128-c8f32694c432/stderr.log  (41101 bytes)","runs/a0e1b107-631d-489d-a128-c8f32694c432/stdout.log  (171 bytes)","runs/a2531baa-439f-4776-a522-baced94a1bf1/","runs/a2531baa-439f-4776-a522-baced94a1bf1/run.json  (391 bytes)","runs/a2531baa-439f-4776-a522-baced94a1bf1/status.json  (96 bytes)","runs/a2531baa-439f-4776-a522-baced94a1bf1/stderr.log  (0 bytes)","runs/a2531baa-439f-4776-a522-baced94a1bf1/stdout.log  (74 bytes)","runs/ce22ffca-f30e-4f4b-bab8-eed096b6ab28/","runs/ce22ffca-f30e-4f4b-bab8-eed096b6ab28/run.json  (393 bytes)","runs/ce22ffca-f30e-4f4b-bab8-eed096b6ab28/status.json  (95 bytes)","runs/ce22ffca-f30e-4f4b-bab8-eed096b6ab28/stderr.log  (48095 bytes)","runs/ce22ffca-f30e-4f4b-bab8-eed096b6ab28/stdout.log  (87 bytes)","runs/e887c62f-bb1d-45af-9fcd-47088bd65bbf/","runs/e887c62f-bb1d-45af-9fcd-47088bd65bbf/run.json  (391 bytes)","runs/e887c62f-bb1d-45af-9fcd-47088bd65bbf/status.json  (95 bytes)","runs/e887c62f-bb1d-45af-9fcd-47088bd65bbf/stderr.log  (0 bytes)","runs/e887c62f-bb1d-45af-9fcd-47088bd65bbf/stdout.log  (74 bytes)","runs/fe07fc39-611b-47f2-b7ca-f19c05d3dc73/","runs/fe07fc39-611b-47f2-b7ca-f19c05d3dc73/run.json  (401 bytes)","runs/fe07fc39-611b-47f2-b7ca-f19c05d3dc73/status.json  (96 bytes)","runs/fe07fc39-611b-47f2-b7ca-f19c05d3dc73/stderr.log  (0 bytes)","runs/fe07fc39-611b-47f2-b7ca-f19c05d3dc73/stdout.log  (382 bytes)"]}},{"tool":"execute_custom_python_analysis","result":{"success":true,"output":"===== code/common.py\n\nimport pandas as pd, numpy as np\ndef load():\n    df = pd.read_csv(\"data/us_macro_quarterly.csv\")\n    df.index = pd.PeriodIndex(df[\"period\"], freq=\"Q\")\n    df = df.iloc[1:].copy()          # first obs of infl/realint is a 0 placeholder (no lag available)\n    df[\"dinfl\"] = df[\"infl\"].diff()\n    df[\"growth\"] = np.log(df[\"realgdp\"]).diff()*400   # real GDP growth, annualised\n    return df\n\n===== code/evalfc.py\n\nimport numpy as np, pandas as pd\nfrom scipy import stats\ndef load_all(window):\n    import glob\n    lin = pd.read_csv(f\"artifacts/forecasts/linear_{window}.csv\")\n    ms  = pd.read_csv(f\"artifacts/forecasts/ms_{window}.csv\")\n    uc  = pd.read_csv(\"artifacts/forecasts/ucsv_g0.2.csv\")\n    uc1 = pd.read_csv(\"artifacts/forecasts/ucsv_g0.1.csv\"); uc1[\"model\"]=\"UCSV_g01\"\n    x = pd.concat([lin,ms,uc,uc1]); x[\"origin\"]=pd.PeriodIndex(x[\"origin\"],freq=\"Q\")\n    w = x.pivot_table(index=[\"h\",\"origin\"],columns=\"model\",values=\"fc\"); a = x.groupby([\"h\",\"origin\"])[\"actual\"].first()\n    w[\"actual\"]=a\n    return w\ndef add_comb(w, members=(\"AR\",\"PC_ut\",\"VAR3\",\"MS_AR\",\"UCSV\")):\n    w=w.copy(); w[\"Comb\"]=w[list(members)].mean(axis=1, skipna=False); return w\ndef dm_test(e1,e2,h,loss=\"se\"):\n    \"\"\"HLN-corrected Diebold-Mariano; H0 equal accuracy; d = L(e1)-L(e2); negative => model1 better.\"\"\"\n    d = (e1**2-e2**2) if loss==\"se\" else (np.abs(e1)-np.abs(e2))\n    d = np.asarray(d); n=len(d); dbar=d.mean()\n    dc = d-dbar; g0=dc@dc/n; s=g0\n    for k in range(1,h):\n        s += 2*(1-k/h)*(dc[k:]@dc[:-k]/n)\n    var=max(s/n,1e-12); stat=dbar/np.sqrt(var)\n    hln=np.sqrt((n+1-2*h+h*(h-1)/n)/n); stat*=hln\n    p=2*stats.t.sf(abs(stat),n-1); return stat,p\ndef cw_test(y,f_small,f_big,h):\n    \"\"\"Clark-West (2007) MSPE-adjusted, one-sided: H0 small model adequate.\"\"\"\n    e1=y-f_small; e2=y-f_big; adj = e1**2-(e2**2-(f_small-f_big)**2)\n    adj=np.asarray(adj); n=len(adj); m=adj.mean(); dc=adj-m; s=dc@dc/n\n    for k in range(1,h): s+=2*(1-k/h)*(dc[k:]@dc[:-k]/n)\n    stat=m/np.sqrt(s/n); return stat, stats.norm.sf(stat)\n\n===== code/fc_ucsv.py\n\nimport sys; sys.path.insert(0,\"code\")\nfrom fcsuite import *\ngamma = float(sys.argv[1]); tag=f\"ucsv_g{sys.argv[1]}\"\nN = 40000; rng = np.random.default_rng(12345)\nd = load(); n=len(d); y = d[\"infl\"].values\ntau = rng.normal(y[:4].mean(), 3.0, N)\nh = rng.normal(np.log(np.var(y[:20])*0.8), 1.0, N)\ng = rng.normal(np.log(0.3), 1.0, N)\ntaus=np.zeros(n); ess=np.zeros(n)\nfor t in range(n):\n    h = h + gamma*rng.standard_normal(N); g = g + gamma*rng.standard_normal(N)\n    tau = tau + np.exp(g/2)*rng.standard_normal(N)\n    lw = -0.5*h - 0.5*(y[t]-tau)**2*np.exp(-h)\n    w = np.exp(lw-lw.max()); w/=w.sum()\n    ess[t]=1/np.sum(w**2)\n    taus[t]=np.sum(w*tau)\n    idx = rng.choice(N,N,p=w)             # multinomial resampling every period\n    tau,h,g = tau[idx],h[idx],g[idx]\nrows=[]\nfor hh in H_LIST:\n    yt = target(d,hh).values\n    for T in origins(d):\n        if T+hh>n-1: continue\n        rows.append((str(d.index[T]),hh,\"UCSV\",taus[T],yt[T]))\nout=pd.DataFrame(rows,columns=[\"origin\",\"h\",\"model\",\"fc\",\"actual\"]); out.to_csv(f\"artifacts/forecasts/{tag}.csv\",index=False)\npd.DataFrame({\"period\":d.index.astype(str),\"tau\":taus,\"ess\":ess}).to_csv(f\"artifacts/forecasts/{tag}_tau.csv\",index=False)\nprint(tag,\"min ESS\",ess.min().round(0)); print(out.groupby(\"h\").apply(lambda g_: np.sqrt(np.mean((g_.fc-g_.actual)**2))).round(3))\n\n===== code/fcsuite.py\n\n\"\"\"Forecasting suite for US inflation. All forecasts are pseudo-out-of-sample:\nat origin t only data <= t is used (direct regressions use pairs (X_s, y_{s+h}) with s+h <= t).\"\"\"\nimport numpy as np, pandas as pd, warnings, sys\nsys.path.insert(0, \"code\")\nfrom common import load\nwarnings.filterwarnings(\"ignore\")\n\nFIRST_ORIGIN = pd.Period(\"1979Q4\", \"Q\")     # first target = 1980Q1\nH_LIST = [1, 4, 8]\n\ndef target(d, h):\n    \"\"\"y_h[t] = mean annualised inflation over t+1..t+h (aligned at origin t).\"\"\"\n    return d[\"infl\"][::-1].rolling(h).mean()[::-1].shift(-1)\n\ndef lagmat(x, p):\n    return np.column_stack([x.shift(k).values for k in range(p)])\n\ndef ols_fit_pred(X, y, xnew):\n    b, *_ = np.linalg.lstsq(X, y, rcond=None)\n    res = y - X @ b\n    return float(xnew @ b), res, b\n\ndef bic_ols(X, y):\n    b, *_ = np.linalg.lstsq(X, y, rcond=None)\n    r = y - X @ b; n = len(y)\n    return n*np.log(np.mean(r**2)) + X.shape[1]*np.log(n)\n\ndef window_bounds(T, h, window, W):\n    \"\"\"positions s of training pairs for origin position T.\"\"\"\n    hi = T - h\n    lo = 0 if window == \"expanding\" else max(0, T - W + 1)\n    return lo, hi\n\ndef origins(d):\n    T0 = d.index.get_loc(FIRST_ORIGIN)\n    return list(range(T0, len(d) - 1))      # origins with at least 1 realised target\n\nartifacts/forecasts/ucsv_g0.2.csv (347, 5)\n   origin  h model         fc  actual\n0  1979Q4  1  UCSV  13.488771   14.60\n1  1980Q1  1  UCSV  14.132742    8.32\n2  1980Q2  1  UCSV  10.741770   10.04\n     origin  h model        fc   actual\n345  2007Q2  8  UCSV  2.837425  1.69250\n346  2007Q3  8  UCSV  2.909811  1.70625\nartifacts/forecasts/linear_expanding.csv (3123, 5)\n   origin  h model        fc  actual\n0  1979Q4  1   RW1  14.62000    14.6\n1  1979Q4  1   RW4  12.98750    14.6\n2  1979Q4  1    AR  12.96518    14.6\n      origin  h  model       fc   actual\n3121  2007Q3  8  IMA11  3.17242  1.70625\n3122  2007Q3  8   VAR3  3.55187  1.70625\nartifacts/forecasts/ucsv_g0.1.csv (347, 5)\n   origin  h model         fc  actual\n0  1979Q4  1  UCSV  13.195278   14.60\n1  1980Q1  1  UCSV  13.939708    8.32\n2  1980Q2  1  UCSV  11.147643   10.04\n     origin  h model        fc   actual\n345  2007Q2  8  UCSV  2.929675  1.69250\n346  2007Q3  8  UCSV  3.025012  1.70625\n\n[Notebook session — variables in memory: f: str, d: DataFrame 347x5]","result":null,"figures":[],"files":[]}},{"tool":"execute_custom_python_analysis","result":{"success":true,"output":"mu  sim_5%  GR2010_5%  sim_10%  GR2010_10%\n0  0.1   3.465      3.393    3.229       3.170\n1  0.2   3.235      3.179    2.976       2.948\n2  0.3   3.080      3.012    2.802       2.766\n3  0.4   2.936      2.890    2.662       2.626\n4  0.5   2.810      2.779    2.525       2.500\n5  0.6   2.688      2.634    2.392       2.356\n6  0.7   2.567      2.475    2.268       2.206\n1.5323939323425293\n\n[Notebook session — variables in memory: f: str, d: DataFrame 347x5, code: str, hac_sigma(), gr_stat(), sim_cv(), t0: float, paper: dict[7], rows: list[7], mu: float, c: dict[2], cv: DataFrame 7x5]","result":null,"figures":[],"files":[]}},{"tool":"execute_custom_python_analysis","result":{"success":true,"output":"116 1979Q4 2008Q3\nrel RMSFE 0.9259956836248732 DM (np.float64(-1.1461339190851916), np.float64(0.25411904261087453))\n{1: 115}\n\n[Notebook session — variables in memory: f: str, d: DataFrame 347x5, code: str, hac_sigma(), gr_stat(), sim_cv(), t0: float, paper: dict[7], rows: list[7], mu: float, c: dict[2], cv: DataFrame 7x5, dm_test(), lin: DataFrame 3123x5, uc: DataFrame 347x5, uc1: DataFrame 347x5, pair(), j: DataFrame 116x3, eA: Series 116, eU: Series 116]","result":null,"figures":[],"files":[]}},{"tool":"execute_custom_python_analysis","result":{"success":true,"output":"29 0.25 {0.05: np.float64(3.155), 0.1: np.float64(2.895)}\n35 0.302 {0.05: np.float64(3.07), 0.1: np.float64(2.809)}\n46 0.397 {0.05: np.float64(2.929), 0.1: np.float64(2.665)}\n58 0.5 {0.05: np.float64(2.811), 0.1: np.float64(2.523)}\n   loss       ucsv   m     mu  L   maxF   minF  absmax    cv5   cv10   rej5  rej10  argmax  share_pos\n0    se  UCSV g=.2  29  0.250  3  2.652 -0.842   2.652  3.155  2.895  False  False  2005Q4      0.852\n1    se  UCSV g=.2  29  0.250  4  2.608 -0.828   2.608  3.155  2.895  False  False  2005Q4      0.852\n2    se  UCSV g=.2  29  0.250  8  2.471 -0.785   2.471  3.155  2.895  False  False  2005Q4      0.852\n3    se  UCSV g=.2  35  0.302  3  2.559 -0.817   2.559  3.070  2.809  False  False  2005Q1      0.878\n4    se  UCSV g=.2  35  0.302  4  2.517 -0.803   2.517  3.070  2.809  False  False  2005Q1      0.878\n5    se  UCSV g=.2  35  0.302  8  2.385 -0.761   2.385  3.070  2.809  False  False  2005Q1      0.878\n6    se  UCSV g=.2  46  0.397  3  2.162 -0.499   2.162  2.929  2.665  False  False  2003Q4      0.901\n7    se  UCSV g=.2  46  0.397  4  2.126 -0.491   2.126  2.929  2.665  False  False  2003Q4      0.901\n8    se  UCSV g=.2  46  0.397  8  2.014 -0.465   2.014  2.929  2.665  False  False  2003Q4      0.901\n9    se  UCSV g=.2  58  0.500  3  2.110 -0.365   2.110  2.811  2.523  False  False  2002Q2      0.881\n10   se  UCSV g=.2  58  0.500  4  2.075 -0.359   2.075  2.811  2.523  False  False  2002Q2      0.881\n11   se  UCSV g=.2  58  0.500  8  1.966 -0.340   1.966  2.811  2.523  False  False  2002Q2      0.881\n12   ae  UCSV g=.2  29  0.250  3  1.858 -0.899   1.858  3.155  2.895  False  False  1995Q2      0.739\n13   ae  UCSV g=.2  29  0.250  4  1.846 -0.893   1.846  3.155  2.895  False  False  1995Q2      0.739\n14   ae  UCSV g=.2  29  0.250  8  1.850 -0.895   1.850  3.155  2.895  False  False  1995Q2      0.739\n15   ae  UCSV g=.2  35  0.302  3  2.007 -1.048   2.007  3.070  2.809  False  False  1994Q3      0.878\n16   ae  UCSV g=.2  35  0.302  4  1.994 -1.041   1.994  3.070  2.809  False  False  1994Q3      0.878\n17   ae  UCSV g=.2  35  0.302  8  1.998 -1.043   1.998  3.070  2.809  False  False  1994Q3      0.878\n18   ae  UCSV g=.2  46  0.397  3  1.514 -0.670   1.514  2.929  2.665  False  False  1998Q3      0.901\n19   ae  UCSV g=.2  46  0.397  4  1.504 -0.666   1.504  2.929  2.665  False  False  1998Q3      0.901\n20   ae  UCSV g=.2  46  0.397  8  1.507 -0.667   1.507  2.929  2.665  False  False  1998Q3      0.901\n21   ae  UCSV g=.2  58  0.500  3  1.684 -0.342   1.684  2.811  2.523  False  False  2002Q2      0.966\n22   ae  UCSV g=.2  58  0.500  4  1.673 -0.340   1.673  2.811  2.523  False  False  2002Q2      0.966\n23   ae  UCSV g=.2  58  0.500  8  1.676 -0.341   1.676  2.811  2.523  False  False  2002Q2      0.966\n24   se  UCSV g=.1  29  0.250  3  2.220 -1.050   2.220  3.155  2.895  False  False  2005Q4      0.852\n25   se  UCSV g=.1  29  0.250  4  2.169 -1.026   2.169  3.155  2.895  False  False  2005Q4      0.852\n26   se  UCSV g=.1  29  0.250  8  2.053 -0.971   2.053  3.155  2.895  False  False  2005Q4      0.852\n27   se  UCSV g=.1  35  0.302  3  2.203 -1.017   2.203  3.070  2.809  False  False  2005Q1      0.890\n28   se  UCSV g=.1  35  0.302  4  2.153 -0.994   2.153  3.070  2.809  False  False  2005Q1      0.890\n29   se  UCSV g=.1  35  0.302  8  2.037 -0.941   2.037  3.070  2.809  False  False  2005Q1      0.890\n30   se  UCSV g=.1  46  0.397  3  1.837 -0.667   1.837  2.929  2.665  False  False  2003Q4      0.901\n31   se  UCSV g=.1  46  0.397  4  1.795 -0.651   1.795  2.929  2.665  False  False  2003Q4      0.901\n32   se  UCSV g=.1  46  0.397  8  1.699 -0.616   1.699  2.929  2.665  False  False  2003Q4      0.901\n33   se  UCSV g=.1  58  0.500  3  1.872 -0.470   1.872  2.811  2.523  False  False  2002Q2      0.881\n34   se  UCSV g=.1  58  0.500  4  1.829 -0.459   1.829  2.811  2.523  False  False  2002Q2      0.881\n35   se  UCSV g=.1  58  0.500  8  1.731 -0.434   1.731  2.811  2.523  False  False  2002Q2      0.881\n\n[Notebook session — variables in memory: f: str, d: ndarray (116,), code: str, hac_sigma(), gr_stat(), sim_cv(), t0: float, paper: dict[7], rows: list[7], mu: float, c: ndarray (59,), cv: DataFrame 7x5, dm_test(), lin: DataFrame 3123x5, uc: DataFrame 347x5, uc1: DataFrame 347x5, pair(), j: DataFrame 116x3, eA: Series 116, eU: Series 116, P: int, dL(), CV: dict[4], m: int, res: list[36], specs: list[3], loss: str, ucdf: DataFrame 347x5, lab: str, jj: DataFrame 116x3, L: int, F: ndarray (59,), sig: float64, R: DataFrame 36x14]","result":null,"figures":[],"files":[]}},{"tool":"execute_custom_python_analysis","result":{"success":true,"output":"m  L  size_5pct_iid_MA3  size_5pct_MA3_heterosk\n0  35  3            0.11425                 0.15050\n1  35  4            0.08100                 0.11700\n2  35  8            0.05000                 0.08125\n3  46  3            0.12100                 0.14550\n4  46  4            0.08975                 0.11400\n5  46  8            0.05700                 0.08600\n[(0.0, np.float64(0.082)), (0.25, np.float64(0.312)), (0.5, np.float64(0.804)), (0.75, np.float64(0.982))]\n\n[Notebook session — variables in memory: f: str, d: ndarray (116,), code: str, hac_sigma(), gr_stat(), sim_cv(), t0: float, paper: dict[7], rows: list[7], mu: float, c: ndarray (59,), cv: dict[2], dm_test(), lin: DataFrame 3123x5, uc: DataFrame 347x5, uc1: DataFrame 347x5, pair(), j: DataFrame 116x3, eA: Series 116, eU: Series 116, P: int, dL(), CV: dict[4], m: int, res: list[36], specs: list[3], loss: str, ucdf: DataFrame 347x5, lab: str, jj: DataFrame 116x3, L: int, F: ndarray (59,), sig: float64, R: DataFrame 36x14, rng: Generator, reps: int, sim(), out: list[6], size: DataFrame 6x4, pw: list[4]]","result":null,"figures":[],"files":[]}},{"tool":"execute_custom_python_analysis","result":{"success":true,"output":"2.3846135914347917 -0.7610393238645863 2005-03-31 23:59:59.999999 2.0143800824737226 2.9286867082639176 54.929521318331766 57.71158450296146 -30.183675068169503 1983-06-30 23:59:59.999999 2009-06-30 23:59:59.999999\n\n---\nFIGURES PRODUCED BY THIS CALL (1) — ready to place in the report:\n\nFigure 1: ![Graphique](/figures/fig-1795213812241-b6fz2h3ys-0.png)\n\n---\nPlace every figure you keep in the section it supports, using its markdown exactly as written above (same URL; the text in brackets is the caption shown under the image, so you may rewrite it as a short descriptive title, without the word \"Figure\" or a number: figures are numbered automatically in reading order). Announce the figure in the sentence before it, and follow it with a paragraph that reads it: what it shows, the number that matters, what it implies. A figure that does not serve the argument is left out rather than dropped at the end.\n\n[Figure check — VQuant visualization standard]\nFigure 1 (“UCSV contre AR à h=4 : un avantage tardif, sans instabilité ”) — to fix in your code: The title or subtitle (“Source : Prévisions pseudo-hors-échantil”) sits on the plot area or its tick labels: leave room above the axes (for example fig.subplots_adjust(top=0.82)) or call tight_layout before placing the title.\nRedraw the figures that still have defects before using them in the report.\n[Vector copies (SVG, PDF) and the PNG of each figure are saved under artifacts/figures/ in the workspace.]\n\n[Notebook session — variables in memory: f: str, d: ndarray (116,), code: str, hac_sigma(), gr_stat(), sim_cv(), t0: float, paper: dict[7], rows: list[7], mu: float, c: ndarray (59,), cv: dict[2], dm_test(), lin: DataFrame 3123x5, uc: DataFrame 347x5, uc1: DataFrame 347x5, pair(), j: DataFrame 116x3, eA: Series 116, eU: Series 116, P: int, dL(), CV: dict[4], m: int, res: list[36], specs: list[3], loss: str, ucdf: DataFrame 347x5, lab: str, jj: DataFrame 116x3, L: int, F: ndarray (59,), sig: float64, R: DataFrame 36x14, rng: Generator, reps: int, sim(), out: list[6], size: DataFrame 6x4, pw: list[4]]","result":null,"files":[]}},{"tool":"execute_custom_python_analysis","result":{"success":true,"output":"{1980: -25.3, 1985: 4.0, 1990: 9.6, 1995: 6.7, 2000: 1.3, 2005: 58.6}\n<Figure size 1000x840 with 2 Axes>\n\n[Notebook session — variables in memory: f: str, d: ndarray (116,), code: str, hac_sigma(), gr_stat(), sim_cv(), t0: float, paper: dict[7], rows: list[7], mu: float, c: ndarray (59,), cv: dict[2], dm_test(), lin: DataFrame 3123x5, uc: DataFrame 347x5, uc1: DataFrame 347x5, pair(), j: DataFrame 116x3, eA: Series 116, eU: Series 116, P: int, dL(), CV: dict[4], m: int, res: list[36], specs: list[3], loss: str, ucdf: DataFrame 347x5, lab: str, jj: DataFrame 116x3, L: int, F: ndarray (59,), sig: float64, R: DataFrame 36x14, rng: Generator, reps: int, sim(), out: list[6], size: DataFrame 6x4, pw: list[4]]","result":null,"figures":[],"files":[]}},{"tool":"execute_custom_python_analysis","result":{"success":true,"output":"\n\n---\nFIGURES PRODUCED BY THIS CALL (1) — ready to place in the report:\n\nFigure 2: ![Graphique](/figures/fig-1795213825486-233hol16p-0.png)\n\n---\nPlace every figure you keep in the section it supports, using its markdown exactly as written above (same URL; the text in brackets is the caption shown under the image, so you may rewrite it as a short descriptive title, without the word \"Figure\" or a number: figures are numbered automatically in reading order). Announce the figure in the sentence before it, and follow it with a paragraph that reads it: what it shows, the number that matters, what it implies. A figure that does not serve the argument is left out rather than dropped at the end.\n\n[Figure check — VQuant visualization standard]\nFigure 1 (“UCSV contre AR à h=4 : un avantage concentré en 2005–2009, s”) — to fix in your code: The title or subtitle (“Source : Prévisions pseudo-hors-échantil”) sits on the plot area or its tick labels: leave room above the axes (for example fig.subplots_adjust(top=0.82)) or call tight_layout before placing the title.\nRedraw the figures that still have defects before using them in the report.\n[Vector copies (SVG, PDF) and the PNG of each figure are saved under artifacts/figures/ in the workspace.]\n\n[Notebook session — variables in memory: f: str, d: ndarray (116,), code: str, hac_sigma(), gr_stat(), sim_cv(), t0: float, paper: dict[7], rows: list[7], mu: float, c: ndarray (59,), cv: dict[2], dm_test(), lin: DataFrame 3123x5, uc: DataFrame 347x5, uc1: DataFrame 347x5, pair(), j: DataFrame 116x3, eA: Series 116, eU: Series 116, P: int, dL(), CV: dict[4], m: int, res: list[36], specs: list[3], loss: str, ucdf: DataFrame 347x5, lab: str, jj: DataFrame 116x3, L: int, F: ndarray (59,), sig: float64, R: DataFrame 36x14, rng: Generator, reps: int, sim(), out: list[6], size: DataFrame 6x4, pw: list[4]]","result":null,"files":[]}},{"tool":"execute_custom_python_analysis","result":{"success":true,"output":"\n\n---\nFIGURES PRODUCED BY THIS CALL (1) — ready to place in the report:\n\nFigure 3: ![Graphique](/figures/fig-1795213833911-rhz2rwtay-0.png)\n\n---\nPlace every figure you keep in the section it supports, using its markdown exactly as written above (same URL; the text in brackets is the caption shown under the image, so you may rewrite it as a short descriptive title, without the word \"Figure\" or a number: figures are numbered automatically in reading order). Announce the figure in the sentence before it, and follow it with a paragraph that reads it: what it shows, the number that matters, what it implies. A figure that does not serve the argument is left out rather than dropped at the end.\n\n[Figure check — VQuant visualization standard]\nFigure 1 (“UCSV contre AR (h=4) : gain concentré en 2005–2009”) — to fix in your code: The title or subtitle (“Source : Prévisions pseudo-hors-échantil”) sits on the plot area or its tick labels: leave room above the axes (for example fig.subplots_adjust(top=0.82)) or call tight_layout before placing the title.\nRedraw the figures that still have defects before using them in the report.\n[Vector copies (SVG, PDF) and the PNG of each figure are saved under artifacts/figures/ in the workspace.]\n\n[Notebook session — variables in memory: f: str, d: ndarray (116,), code: str, hac_sigma(), gr_stat(), sim_cv(), t0: float, paper: dict[7], rows: list[7], mu: float, c: ndarray (59,), cv: dict[2], dm_test(), lin: DataFrame 3123x5, uc: DataFrame 347x5, uc1: DataFrame 347x5, pair(), j: DataFrame 116x3, eA: Series 116, eU: Series 116, P: int, dL(), CV: dict[4], m: int, res: list[36], specs: list[3], loss: str, ucdf: DataFrame 347x5, lab: str, jj: DataFrame 116x3, L: int, F: ndarray (59,), sig: float64, R: DataFrame 36x14, rng: Generator, reps: int, sim(), out: list[6], size: DataFrame 6x4, pw: list[4]]","result":null,"files":[]}},{"tool":"execute_custom_python_analysis","result":{"success":true,"output":"<Figure size 1000x900 with 2 Axes>\n\n[Notebook session — variables in memory: f: str, d: ndarray (116,), code: str, hac_sigma(), gr_stat(), sim_cv(), t0: float, paper: dict[7], rows: list[7], mu: float, c: ndarray (59,), cv: dict[2], dm_test(), lin: DataFrame 3123x5, uc: DataFrame 347x5, uc1: DataFrame 347x5, pair(), j: DataFrame 116x3, eA: Series 116, eU: Series 116, P: int, dL(), CV: dict[4], m: int, res: list[36], specs: list[3], loss: str, ucdf: DataFrame 347x5, lab: str, jj: DataFrame 116x3, L: int, F: ndarray (59,), sig: float64, R: DataFrame 36x14, rng: Generator, reps: int, sim(), out: list[6], size: DataFrame 6x4, pw: list[4]]","result":null,"figures":[],"files":[]}},{"tool":"execute_custom_python_analysis","result":{"success":true,"output":"\n\n---\nFIGURES PRODUCED BY THIS CALL (1) — ready to place in the report:\n\nFigure 4: ![Graphique](/figures/fig-1795213844205-hhyyduslm-0.png)\n\n---\nPlace every figure you keep in the section it supports, using its markdown exactly as written above (same URL; the text in brackets is the caption shown under the image, so you may rewrite it as a short descriptive title, without the word \"Figure\" or a number: figures are numbered automatically in reading order). Announce the figure in the sentence before it, and follow it with a paragraph that reads it: what it shows, the number that matters, what it implies. A figure that does not serve the argument is left out rather than dropped at the end.\n\n[Figure check — VQuant visualization standard]\nFigure 1 (“UCSV contre AR (h=4) : gain concentré en 2005–2009”) — to fix in your code: The title or subtitle (“Source : Prévisions pseudo-hors-échantil”) sits on the plot area or its tick labels: leave room above the axes (for example fig.subplots_adjust(top=0.82)) or call tight_layout before placing the title.\nRedraw the figures that still have defects before using them in the report.\n[Vector copies (SVG, PDF) and the PNG of each figure are saved under artifacts/figures/ in the workspace.]\n\n[Notebook session — variables in memory: f: str, d: ndarray (116,), code: str, hac_sigma(), gr_stat(), sim_cv(), t0: float, paper: dict[7], rows: list[7], mu: float, c: ndarray (59,), cv: dict[2], dm_test(), lin: DataFrame 3123x5, uc: DataFrame 347x5, uc1: DataFrame 347x5, pair(), j: DataFrame 116x3, eA: Series 116, eU: Series 116, P: int, dL(), CV: dict[4], m: int, res: list[36], specs: list[3], loss: str, ucdf: DataFrame 347x5, lab: str, jj: DataFrame 116x3, L: int, F: ndarray (59,), sig: float64, R: DataFrame 36x14, rng: Generator, reps: int, sim(), out: list[6], size: DataFrame 6x4, pw: list[4]]","result":null,"files":[]}},{"tool":"execute_custom_python_analysis","result":{"success":true,"output":"'UCSV contre AR (h=4) : gain concentré en 2005–200 [100, 878, 651, 898] True\n'Haut : gain cumulé en erreur quadratique ; bas :  [100, 814, 805, 828] True\n'Source : Prévisions pseudo-hors-échantillon 1980T [100, 363, 539, 376] True\n'vquant' [952, 4, 995, 16] True\n'courbe qui descend : AR meilleur\\ncourbe qui mont [399, 527, 622, 557] False\n'2005–2009 :\\n+58,6 sur +54,9' [687, 527, 793, 557] False\n'' [535, 796, 535, 796] False\nax0 bbox [100, 502, 970, 774] fig px [1000.  900.]\n\n[Notebook session — variables in memory: f: str, d: ndarray (116,), code: str, hac_sigma(), gr_stat(), sim_cv(), t0: float, paper: dict[7], rows: list[7], mu: float, c: ndarray (59,), cv: dict[2], dm_test(), lin: DataFrame 3123x5, uc: DataFrame 347x5, uc1: DataFrame 347x5, pair(), j: DataFrame 116x3, eA: Series 116, eU: Series 116, P: int, dL(), CV: dict[4], m: int, res: list[36], specs: list[3], loss: str, ucdf: DataFrame 347x5, lab: str, jj: DataFrame 116x3, L: int, F: ndarray (59,), sig: float64, R: DataFrame 36x14, rng: Generator, reps: int, sim(), out: list[6], size: DataFrame 6x4, pw: list[4]]","result":null,"figures":[],"files":[]}},{"tool":"execute_custom_python_analysis","result":{"success":true,"output":"\n\n---\nFIGURES PRODUCED BY THIS CALL (1) — ready to place in the report:\n\nFigure 5: ![Graphique](/figures/fig-1795213857744-yrr6kz9wk-0.png)\n\n---\nPlace every figure you keep in the section it supports, using its markdown exactly as written above (same URL; the text in brackets is the caption shown under the image, so you may rewrite it as a short descriptive title, without the word \"Figure\" or a number: figures are numbered automatically in reading order). Announce the figure in the sentence before it, and follow it with a paragraph that reads it: what it shows, the number that matters, what it implies. A figure that does not serve the argument is left out rather than dropped at the end.\n\n[Figure check — VQuant visualization standard]\nFigure 1 — repaired automatically: contrast.\n[Vector copies (SVG, PDF) and the PNG of each figure are saved under artifacts/figures/ in the workspace.]\n\n[Notebook session — variables in memory: f: str, d: ndarray (116,), code: str, hac_sigma(), gr_stat(), sim_cv(), t0: float, paper: dict[7], rows: list[7], mu: float, c: ndarray (59,), cv: dict[2], dm_test(), lin: DataFrame 3123x5, uc: DataFrame 347x5, uc1: DataFrame 347x5, pair(), j: DataFrame 116x3, eA: Series 116, eU: Series 116, P: int, dL(), CV: dict[4], m: int, res: list[36], specs: list[3], loss: str, ucdf: DataFrame 347x5, lab: str, jj: DataFrame 116x3, L: int, F: ndarray (59,), sig: float64, R: DataFrame 36x14, rng: Generator, reps: int, sim(), out: list[6], size: DataFrame 6x4, pw: list[4]]","result":null,"files":[]}},{"tool":"execute_custom_python_analysis","result":{"success":true,"output":"sans 2005-09: n 100 rel RMSFE 1.0119121726912472 DM (np.float64(0.25479311407181077), np.float64(0.799411424099517))\n2005-09 only: n 16 0.7736747275541955\nRMSFE AR/UCSV full 1.8227106547508494 1.6878221987963529\norigin\n2009-06-30 23:59:59.999999    27.2\n2006-09-30 23:59:59.999999    13.9\n2008-12-31 23:59:59.999999    11.2\n2009-03-31 23:59:59.999999     8.5\n1991-09-30 23:59:59.999999     3.4\n2006-12-31 23:59:59.999999     3.2\ndtype: float64\n752\n\n[Notebook session — variables in memory: f: str, d: ndarray (116,), code: str, hac_sigma(), gr_stat(), sim_cv(), t0: float, paper: dict[7], rows: list[7], mu: float, c: ndarray (59,), cv: dict[2], dm_test(), lin: DataFrame 3123x5, uc: DataFrame 347x5, uc1: DataFrame 347x5, pair(), j: DataFrame 116x3, eA: Series 116, eU: Series 116, P: int, dL(), CV: dict[4], m: int, res: list[36], specs: list[3], loss: str, ucdf: DataFrame 347x5, lab: str, jj: DataFrame 116x3, L: int, F: ndarray (59,), sig: float64, R: DataFrame 36x14, rng: Generator, reps: int, sim(), out: list[6], size: DataFrame 6x4, pw: list[4]]","result":null,"figures":[],"files":[]}},{"tool":"research_run","result":{"run_id":"331159db-1fa0-4a55-a785-2663a77efcec","label":"Vérification reproductible du test GR (script autonome)","code":"code/gr_run.py","status":"done","exit_code":0,"seconds":3,"progress":1,"error":null,"logs":"runs/331159db-1fa0-4a55-a785-2663a77efcec/stdout.log, runs/331159db-1fa0-4a55-a785-2663a77efcec/stderr.log","stdout_tail":"35 max|F| 2.385 cv5 3.07 cv10 2.809\n46 max|F| 2.014 cv5 2.929 cv10 2.665\n","files_written":[]}},{"tool":"log_experiment","result":{"experiment_id":"00d7f21b-1726-4464-b9df-f54d4c489186","number":82,"status":"ok","recorded":true}},{"tool":"research_memory","result":{"saved":"9912a5dd","kind":"finding","title":"Giacomini-Rossi fluctuation test UCSV vs AR h=4 (done)"}}],"sources":[],"searchQueries":[],"pythonCode":null,"monteCarloResults":null,"customPythonFigures":[{"id":"fig-1795213812241-b6fz2h3ys","figures":["iVBORw0KGgoAAAANSUhEUgAAB7cAAAdNCAYAAAAI+97tAAAA…(trimmed 298224 chars)"],"figureUrls":["/figures/fig-1795213812241-b6fz2h3ys-0.png"],"output":"2.3846135914347917 -0.7610393238645863 2005-03-31 23:59:59.999999 2.0143800824737226 2.9286867082639176 54.929521318331766 57.71158450296146 -30.183675068169503 1983-06-30 23:59:59.999999 2009-06-30 23:59:59.999999\n\n---\nFIGURES PRODUCED BY THIS CALL (1) — ready to place in the report:\n\nFigure 1: ![Graphique](/figures/fig-1795213812241-b6fz2h3ys-0.png)\n\n---\nPlace every figure you keep in the section it supports, using its markdown exactly as written above (same URL; the text in brackets is the caption shown under the image, so you may rewrite it as a short descriptive title, without the word \"Figure\" or a number: figures are numbered automatically in reading order). Announce the figure in the sentence before it, and follow it with a paragraph that reads it: what it shows, the number that matters, what it implies. A figure that does not serve the argument is left out rather than dropped at the end."},{"id":"fig-1795213825486-233hol16p","figures":["iVBORw0KGgoAAAANSUhEUgAACGMAAAd2CAYAAACjRPJ2AAAA…(trimmed 318240 chars)"],"figureUrls":["/figures/fig-1795213825486-233hol16p-0.png"],"output":"\n\n---\nFIGURES PRODUCED BY THIS CALL (1) — ready to place in the report:\n\nFigure 2: ![Graphique](/figures/fig-1795213825486-233hol16p-0.png)\n\n---\nPlace every figure you keep in the section it supports, using its markdown exactly as written above (same URL; the text in brackets is the caption shown under the image, so you may rewrite it as a short descriptive title, without the word \"Figure\" or a number: figures are numbered automatically in reading order). Announce the figure in the sentence before it, and follow it with a paragraph that reads it: what it shows, the number that matters, what it implies. A figure that does not serve the argument is left out rather than dropped at the end."},{"id":"fig-1795213833911-rhz2rwtay","figures":["iVBORw0KGgoAAAANSUhEUgAAB+kAAAcTCAYAAACZogGRAAAA…(trimmed 300132 chars)"],"figureUrls":["/figures/fig-1795213833911-rhz2rwtay-0.png"],"output":"\n\n---\nFIGURES PRODUCED BY THIS CALL (1) — ready to place in the report:\n\nFigure 3: ![Graphique](/figures/fig-1795213833911-rhz2rwtay-0.png)\n\n---\nPlace every figure you keep in the section it supports, using its markdown exactly as written above (same URL; the text in brackets is the caption shown under the image, so you may rewrite it as a short descriptive title, without the word \"Figure\" or a number: figures are numbered automatically in reading order). Announce the figure in the sentence before it, and follow it with a paragraph that reads it: what it shows, the number that matters, what it implies. A figure that does not serve the argument is left out rather than dropped at the end."},{"id":"fig-1795213844205-hhyyduslm","figures":["iVBORw0KGgoAAAANSUhEUgAAB+kAAAejCAYAAACDaBLkAAAA…(trimmed 303384 chars)"],"figureUrls":["/figures/fig-1795213844205-hhyyduslm-0.png"],"output":"\n\n---\nFIGURES PRODUCED BY THIS CALL (1) — ready to place in the report:\n\nFigure 4: ![Graphique](/figures/fig-1795213844205-hhyyduslm-0.png)\n\n---\nPlace every figure you keep in the section it supports, using its markdown exactly as written above (same URL; the text in brackets is the caption shown under the image, so you may rewrite it as a short descriptive title, without the word \"Figure\" or a number: figures are numbered automatically in reading order). Announce the figure in the sentence before it, and follow it with a paragraph that reads it: what it shows, the number that matters, what it implies. A figure that does not serve the argument is left out rather than dropped at the end."},{"id":"fig-1795213857744-yrr6kz9wk","figures":["iVBORw0KGgoAAAANSUhEUgAAB+kAAAejCAYAAACDaBLkAAAA…(trimmed 303384 chars)"],"figureUrls":["/figures/fig-1795213857744-yrr6kz9wk-0.png"],"output":"\n\n---\nFIGURES PRODUCED BY THIS CALL (1) — ready to place in the report:\n\nFigure 5: ![Graphique](/figures/fig-1795213857744-yrr6kz9wk-0.png)\n\n---\nPlace every figure you keep in the section it supports, using its markdown exactly as written above (same URL; the text in brackets is the caption shown under the image, so you may rewrite it as a short descriptive title, without the word \"Figure\" or a number: figures are numbered automatically in reading order). Announce the figure in the sentence before it, and follow it with a paragraph that reads it: what it shows, the number that matters, what it implies. A figure that does not serve the argument is left out rather than dropped at the end."}],"provenance":[{"tool":"execute_custom_python_analysis","provider":"python","symbols":[],"retrievedAt":1795213864544,"ok":true,"calls":15},{"tool":"research_run","provider":"python","symbols":[],"retrievedAt":1795213866206,"ok":true,"calls":2}],"model":"claude-sonnet-5-5","generationId":"c776e64f-f227-42c2-ae53-ca7ed1a1becc","usage":{"inputTokens":2039123,"outputTokens":23346,"costUsd":0.753365,"knownCostUsd":0.753365,"byModel":[{"model":"claude-sonnet-5-5","provider":"anthropic","role":"lead","calls":20,"inputTokens":2039123,"outputTokens":23346,"costUsd":0.753365}]},"research":{"workspaceId":"s-3d04acf2-235e-46e3-8c2d-bf26b0ba3c81","experiments":[{"id":"00d7f21b-1726-4464-b9df-f54d4c489186","seq":82,"title":"Test de fluctuation de Giacomini-Rossi (2010), UCSV vs AR, h=4","status":"ok","metrics":{"P":116,"rel_RMSFE":0.926,"DM_p":0.254,"maxF_m35_L8":2.385,"cv5_m35":3.07,"cv10_m35":2.809,"maxF_m46_L8":2.014,"cv5_m46":2.929,"rejets_sur_36_specs":0,"taille_5pct_L4_m35_iid":0.081,"taille_5pct_L8_m35_iid":0.05,"puissance_decalage_0.5sd":0.8,"RMSFE_sans_2005_09_rel":1.012,"DM_p_sans_2005_09":0.8},"error":null}],"executions":{"total":16,"failed":1},"qualityGate":[{"id":"numbers","severity":"warning","finding":"1 of 33 figures quoted in the text are not found in the stored results (within rounding and unit changes). They may be derived in the prose, come from a source that was not saved, or be wrong.","evidence":"\"60 000\" in «la main dans . - **Seuils :** simulés (60 000 réplications) à partir de la»"},{"id":"sample","severity":"info","finding":"Sample size and period are stated for the statistical statements.","evidence":"Stated in the text."}]}}],"createdAt":"2026-11-20T22:24:48.000Z","updatedAt":"2026-11-20T22:31:38.000Z"}
Files main
Menu

Theme