# Literature ledger — WP 27 "The Invisible Hand of Machines"

Claim-to-citation ledger. Every row was verified on 2026-10-27 against Crossref (`https://api.crossref.org/works/<DOI>`), the arXiv API / abstract page, or the publisher page named in the column. Bib keys match `paper/references.bib`. Items that could not be verified are listed at the end and are **not** in the bib file.


## (a) Agent-based computational economics and heterogeneous-agent models

| # | claim id | exact claim we will make in the paper | citation key | DOI or URL | year | how verified | verification date |
|---|---|---|---|---|---|---|---|
| 1 | A001 | Agent-based computational finance reproduces fat tails, volatility clustering and bubbles from interacting heterogeneous rules; survey we build on for the baseline rule-based population. | `lebaron2006` | 10.1016/S1574-0021(05)02024-1 | 2006 | Crossref (title/authors/venue matched) | 2026-10-27 |
| 2 | A002 | The Santa Fe artificial stock market shows that the speed of belief updating governs whether prices track fundamentals or display excess volatility and technical-trading regimes. | `lebaron1999` | 10.1016/S0165-1889(98)00081-5 | 1999 | Crossref (title/authors/venue matched) | 2026-10-27 |
| 3 | A003 | Interacting noise traders switching between optimistic/pessimistic and fundamentalist states generate scaling laws and clustered volatility endogenously. | `lux1999` | 10.1038/17290 | 1999 | Crossref (title/authors/venue matched) | 2026-10-27 |
| 4 | A004 | Volatility clustering arises from endogenous switching between chartist and fundamentalist strategies in a microsimulation. | `lux2000` | 10.1142/S0219024900000826 | 2000 | Crossref (title/authors/venue matched) | 2026-10-27 |
| 5 | A005 | Adaptive belief systems with discrete-choice switching between predictors produce complex (bifurcating) price dynamics as intensity of choice rises. | `brock1997` | 10.2307/2171879 | 1997 | Crossref (title/authors/venue matched) | 2026-10-27 |
| 6 | A006 | Fundamentalist/chartist heterogeneous-beliefs asset-pricing model: the share of trend-followers determines stability of the fundamental price (used to motivate our policy mix). | `brock1998` | 10.1016/S0165-1889(98)00011-6 | 1998 | Crossref (title/authors/venue matched) | 2026-10-27 |
| 7 | A007 | Simulated double-auction microstructure with fundamentalist, chartist and noise traders reproduces order-book stylized facts; template for our exchange design. | `chiarella2002` | 10.1088/1469-7688/2/5/303 | 2002 | Crossref (title/authors/venue matched) | 2026-10-27 |
| 8 | A008 | Heterogeneous trading rules in a limit-order book generate fat tails, volatility clustering and order-flow autocorrelation; chartist share drives excess volatility. | `chiarella2009` | 10.1016/j.jedc.2008.08.001 | 2009 | Crossref (title/authors/venue matched) | 2026-10-27 |
| 9 | A009 | Agent-based models are the appropriate tool for studying out-of-equilibrium market dynamics and policy counterfactuals that cannot be solved analytically. | `farmer2009` | 10.1038/460685a | 2009 | Crossref (title/authors/venue matched) | 2026-10-27 |
| 10 | A010 | Survey of heterogeneous-agent models: strategy switching and bounded rationality explain bubbles, crashes and excess volatility. | `hommes2006` | 10.1016/S1574-0021(05)02023-X | 2006 | Crossref (title/authors/venue matched) | 2026-10-27 |
| 11 | A011 | Complex-systems view of behavioral macro/finance: laboratory and ABM evidence that expectations heterogeneity drives aggregate instability; justifies treating LLM policies as another expectation type. | `hommes2021` | 10.1257/jel.20191434 | 2021 | Crossref (title/authors/venue matched) | 2026-10-27 |
| 12 | A012 | Recruitment/epidemic models of opinion switching generate persistent herding states without any informational reason. | `kirman1993` | 10.2307/2118498 | 1993 | Crossref (title/authors/venue matched) | 2026-10-27 |
| 13 | A013 | Herd behaviour among speculators generates bubbles and crashes in a mean-field model of contagious optimism. | `lux1995` | 10.2307/2235156 | 1995 | Crossref (title/authors/venue matched) | 2026-10-27 |
| 14 | A014 | Current state of agent-based modeling in economics and finance (JEL survey): validation, calibration and scale are the open frontiers our design addresses with matched seeds and power analysis. | `axtell2025` | 10.1257/jel.20221319 | 2025 | Crossref (title/authors/venue matched) | 2026-10-27 |
| 15 | A015 | Structural stochastic volatility models with herding (fundamentalist/chartist switching) can be estimated by matching moments of stylized facts; we use the same moment set for validation. | `franke2012` | 10.1016/j.jedc.2011.10.004 | 2012 | Crossref (title/authors/venue matched) | 2026-10-27 |

## (b) Zero-intelligence, double-auction microstructure and information aggregation

| # | claim id | exact claim we will make in the paper | citation key | DOI or URL | year | how verified | verification date |
|---|---|---|---|---|---|---|---|
| 16 | B016 | Zero-intelligence traders constrained by budgets reach near-full allocative efficiency in double auctions: market structure, not trader rationality, drives efficiency (our ZI baseline). | `gode1993` | 10.1086/261868 | 1993 | Crossref (title/authors/venue matched) | 2026-10-27 |
| 17 | B017 | A zero-intelligence model of the continuous double auction predicts spread and price-impact relations from order-flow rates alone. | `farmer2005` | 10.1073/pnas.0409157102 | 2005 | Crossref (title/authors/venue matched) | 2026-10-27 |
| 18 | B018 | Statistical theory of the continuous double auction: order-flow parameters determine spread, depth and price diffusion in the ZI limit. | `smith2003` | 10.1088/1469-7688/3/6/307 | 2003 | Crossref (title/authors/venue matched) | 2026-10-27 |
| 19 | B019 | Empirical statistics of stock order books (limit-order placement distribution, book shape) used as reference facts for our simulated book. | `bouchaud2002` | 10.1088/1469-7688/2/4/301 | 2002 | Crossref (title/authors/venue matched) | 2026-10-27 |
| 20 | B020 | Stochastic order-book model with Markovian order flow: conditional probabilities of price moves computed from book state; benchmark for our book-state features. | `cont2010` | 10.1287/opre.1090.0780 | 2010 | Crossref (title/authors/venue matched) | 2026-10-27 |
| 21 | B021 | Order-flow imbalance (OFI) linearly predicts short-horizon price changes; we use OFI as a mechanism metric for synchronized LLM order flow. | `cont2014` | 10.1093/jjfinec/nbt003 | 2014 | Crossref (title/authors/venue matched) | 2026-10-27 |
| 22 | B022 | Reference treatment of order-book dynamics, price impact and the slow digestion of order flow; defines the impact and liquidity measures we report. | `bouchaud2018` | 10.1017/9781316659335 | 2018 | Crossref (title/authors/venue matched) | 2026-10-27 |
| 23 | B023 | Survey of limit-order-book models and empirical regularities used to parametrize the exchange. | `gould2013` | 10.1080/14697688.2013.803148 | 2013 | Crossref (title/authors/venue matched) | 2026-10-27 |
| 24 | B024 | Review of zero-intelligence models in economics and finance: ZI benchmarks isolate the contribution of market structure from agent intelligence. | `ladley2012` | 10.1017/s0269888912000173 | 2012 | Crossref (title/authors/venue matched) | 2026-10-27 |
| 25 | B025 | Continuous double auctions with induced values converge to competitive equilibrium with few traders; origin of the experimental double-auction paradigm. | `smith1962` | 10.1086/258609 | 1962 | Crossref (title/authors/venue matched) | 2026-10-27 |
| 26 | B026 | Experimental asset markets with known dividend streams produce bubbles and crashes; our dividend-paying asset and bubble metrics follow this design. | `smith1988` | 10.2307/1911361 | 1988 | Crossref (title/authors/venue matched) | 2026-10-27 |
| 27 | B027 | Insider trading with a risk-neutral market maker: price impact lambda = Cov(v,Y)/Var(Y) and the informativeness of prices is pinned down by informed-order variance relative to noise; our stylized model is a static Kyle variant. | `kyle1985` | 10.2307/1913210 | 1985 | Crossref (title/authors/venue matched) | 2026-10-27 |
| 28 | B028 | Bid-ask spreads arise from adverse selection when some traders are informed; informed share determines the spread. | `glosten1985` | 10.1016/0304-405X(85)90044-3 | 1985 | Crossref (title/authors/venue matched) | 2026-10-27 |
| 29 | B029 | Competitive noisy rational-expectations equilibrium: prices aggregate dispersed private signals imperfectly because of noise supply; benchmark for informativeness. | `hellwig1980` | 10.1016/0022-0531(80)90056-3 | 1980 | Crossref (title/authors/venue matched) | 2026-10-27 |
| 30 | B030 | Prices cannot be fully informationally efficient when information is costly; the equilibrium degree of informativeness is bounded. | `grossman1980` | https://www.jstor.org/stable/1805228 (AER 70(3):393-408) | 1980 | publisher/IDEAS record (ideas.repec.org/a/aea/aecrev/v70y1980i3p393-408.html); no Crossref DOI | 2026-10-27 |
| 31 | B031 | With correlated private signals among strategic informed traders, the correlation structure of signals (not only their precision) governs trading intensity and price informativeness. | `foster1996` | 10.1111/j.1540-6261.1996.tb04075.x | 1996 | Crossref (title/authors/venue matched) | 2026-10-27 |
| 32 | B032 | Imperfect competition among informed traders with correlated signals: more correlated information leads to faster revelation (rat race), with implications for how shared signals affect price discovery. | `back2000` | 10.1111/0022-1082.00282 | 2000 | Crossref (title/authors/venue matched) | 2026-10-27 |
| 33 | B033 | Public information can increase disagreement and trading volume through higher-order beliefs; shared public signals do not trivially homogenize behaviour. | `kondor2012` | 10.1093/restud/rdr051 | 2012 | Crossref (title/authors/venue matched) | 2026-10-27 |
| 34 | B034 | Information diversity (different traders informed about different components) is complementary; loss of diversity reduces price informativeness about some components. | `goldstein2015` | 10.1111/jofi.12226 | 2015 | Crossref (title/authors/venue matched) | 2026-10-27 |
| 35 | B035 | Abundant cheap data can lower price informativeness about fundamentals by crowding out costly deep information; parallel to shared-model crowding. | `dugast2018` | 10.1016/j.jfineco.2018.07.004 | 2018 | Crossref (title/authors/venue matched) | 2026-10-27 |
| 36 | B036 | Realism metrics (stylized facts) for evaluating limit-order-book simulations; adopted as validation checklist. | `vyetrenko2020` | 10.1145/3383455.3422561 | 2020 | Crossref (title/authors/venue matched) | 2026-10-27 |
| 37 | B037 | ABIDES: discrete-event multi-agent market simulator with latency and a NASDAQ-like matching engine; our exchange follows the same event-driven design. | `byrd2020` | 10.1145/3384441.3395986 | 2020 | Crossref (title/authors/venue matched) | 2026-10-27 |

## (c) Price discovery and information-efficiency measurement

| # | claim id | exact claim we will make in the paper | citation key | DOI or URL | year | how verified | verification date |
|---|---|---|---|---|---|---|---|
| 38 | C038 | Information content of trades measured by a VAR of quote revisions and trades (permanent price impact); we use the Hasbrouck impulse-response measure. | `hasbrouck1991` | 10.1111/j.1540-6261.1991.tb03749.x | 1991 | Crossref (title/authors/venue matched) | 2026-10-27 |
| 39 | C039 | Decomposition of transaction price into an efficient (random-walk) component and a pricing error; the standard deviation of the pricing error is our market-quality metric. | `hasbrouck1993` | 10.1093/rfs/6.1.191 | 1993 | Crossref (title/authors/venue matched) | 2026-10-27 |
| 40 | C040 | Information shares attribute price discovery to venues/agents via the common efficient price; adapted to attribute discovery to LLM vs rule-based order flow. | `hasbrouck1995` | 10.1111/j.1540-6261.1995.tb04054.x | 1995 | Crossref (title/authors/venue matched) | 2026-10-27 |
| 41 | C041 | Variance-ratio test rejects random walk under positive autocorrelation; we use variance ratios as an efficiency statistic. | `lo1988` | 10.1093/rfs/1.1.41 | 1988 | Crossref (title/authors/venue matched) | 2026-10-27 |
| 42 | C042 | Effective bid-ask spread can be inferred from the serial covariance of price changes. | `roll1984` | 10.1111/j.1540-6261.1984.tb03897.x | 1984 | Crossref (title/authors/venue matched) | 2026-10-27 |
| 43 | C043 | Illiquidity measured by absolute return per unit of volume; used as our price-impact proxy. | `amihud2002` | 10.1016/S1386-4181(01)00024-6 | 2002 | Crossref (title/authors/venue matched) | 2026-10-27 |
| 44 | C044 | Price informativeness is measured by how well prices predict future fundamentals (forecasting-regression R2); we use this to define informativeness about the latent dividend process. | `bai2016` | 10.1016/j.jfineco.2016.08.005 | 2016 | Crossref (title/authors/venue matched) | 2026-10-27 |
| 45 | C045 | Price informativeness matters for real decisions (feedback effects), motivating informativeness as a welfare-relevant outcome. | `bond2012` | 10.1146/annurev-financial-110311-101826 | 2012 | Crossref (title/authors/venue matched) | 2026-10-27 |
| 46 | C046 | Transitory price pressures from inventory imbalances are distinguished from permanent information effects; justifies separating pricing error from efficient-price variance. | `hendershott2014` | 10.1016/j.jfineco.2014.08.001 | 2014 | Crossref (title/authors/venue matched) | 2026-10-27 |

## (d) Liquidity spirals, fire sales and margin

| # | claim id | exact claim we will make in the paper | citation key | DOI or URL | year | how verified | verification date |
|---|---|---|---|---|---|---|---|
| 47 | D047 | Funding liquidity and market liquidity reinforce each other through margins, producing liquidity spirals; our margin-call mechanism follows this logic. | `brunnermeier2009` | 10.1093/rfs/hhn098 | 2009 | Crossref (title/authors/venue matched) | 2026-10-27 |
| 48 | D048 | Liquidation values fall when potential buyers are themselves constrained; fire-sale discounts emerge from industry-wide distress. | `shleifer1992` | 10.1111/j.1540-6261.1992.tb04661.x | 1992 | Crossref (title/authors/venue matched) | 2026-10-27 |
| 49 | D049 | Limits of arbitrage: constrained arbitrageurs can amplify rather than correct mispricing when losses trigger withdrawals. | `shleifer1997` | 10.1111/j.1540-6261.1997.tb03807.x | 1997 | Crossref (title/authors/venue matched) | 2026-10-27 |
| 50 | D050 | Fire sales in finance and macroeconomics: forced sales by constrained agents push prices below fundamentals and propagate losses. | `shleifer2011` | 10.1257/jep.25.1.29 | 2011 | Crossref (title/authors/venue matched) | 2026-10-27 |
| 51 | D051 | The leverage cycle: endogenous margins amplify booms and crashes; leverage caps are a policy lever. | `geanakoplos2010` | 10.1086/648285 | 2010 | Crossref (title/authors/venue matched) | 2026-10-27 |
| 52 | D052 | Procyclical leverage of intermediaries: balance-sheet adjustment amplifies price moves. | `adrian2010` | 10.1016/j.jfi.2008.12.002 | 2010 | Crossref (title/authors/venue matched) | 2026-10-27 |
| 53 | D053 | Financially constrained arbitrageurs may under-provide liquidity; welfare effects of constraints. | `gromb2002` | 10.1016/S0304-405X(02)00228-3 | 2002 | Crossref (title/authors/venue matched) | 2026-10-27 |
| 54 | D054 | Leverage in an agent-based model generates fat tails and clustered volatility via synchronized margin-driven selling; closest ABM antecedent of our leverage treatment. | `thurner2012` | 10.1080/14697688.2012.674301 | 2012 | Crossref (title/authors/venue matched) | 2026-10-27 |
| 55 | D055 | Contagion as a wealth effect: convergence traders' losses in one asset propagate to others through reduced risk-bearing capacity. | `kyle2001` | 10.1111/0022-1082.00373 | 2001 | Crossref (title/authors/venue matched) | 2026-10-27 |
| 56 | D056 | Distressed selling by funds induces endogenous correlation between assets; multi-asset spillover mechanism in our K>1 runs. | `cont2013` | 10.1111/j.1467-9965.2011.00510.x | 2013 | Crossref (title/authors/venue matched) | 2026-10-27 |
| 57 | D057 | Fire-sale spillovers across institutions scale with common exposures and leverage; aggregate vulnerability index. | `greenwood2015` | 10.1016/j.jfineco.2014.11.006 | 2015 | Crossref (title/authors/venue matched) | 2026-10-27 |

## (e) Herding, correlated trading, algorithmic trading and flash crashes

| # | claim id | exact claim we will make in the paper | citation key | DOI or URL | year | how verified | verification date |
|---|---|---|---|---|---|---|---|
| 58 | E058 | Herding of institutional investors is measurable by the buy-side imbalance statistic (LSV measure) which we compute for LLM cohorts. | `lakonishok1992` | 10.1016/0304-405X(92)90023-Q | 1992 | Crossref (title/authors/venue matched) | 2026-10-27 |
| 59 | E059 | During the Flash Crash, HFTs did not trigger the crash but their synchronized inventory management amplified it; crash-amplification channel. | `kirilenko2017` | 10.1111/jofi.12498 | 2017 | Crossref (title/authors/venue matched) | 2026-10-27 |
| 60 | E060 | High-frequency market makers behave as new market makers with tight inventory control; informs our market-maker policy. | `menkveld2013` | 10.1016/j.finmar.2013.06.006 | 2013 | Crossref (title/authors/venue matched) | 2026-10-27 |
| 61 | E061 | Algorithmic trading improves liquidity on average (causal evidence from NYSE autoquote). | `hendershott2011` | 10.1111/j.1540-6261.2010.01624.x | 2011 | Crossref (title/authors/venue matched) | 2026-10-27 |
| 62 | E062 | Continuous-time trading creates a latency arms race with correlated sniping; frequent batch auctions as alternative design (motivates our latency and batching treatments). | `budish2015` | 10.1093/qje/qjv027 | 2015 | Crossref (title/authors/venue matched) | 2026-10-27 |
| 63 | E063 | Informational cascades: rational agents ignore private signals after observing predecessors; cascade probability rises with signal correlation. | `bikhchandani1992` | 10.1086/261849 | 1992 | Crossref (title/authors/venue matched) | 2026-10-27 |
| 64 | E064 | Simple model of herd behavior: sequential decision-makers follow others, destroying information aggregation. | `banerjee1992` | 10.2307/2118364 | 1992 | Crossref (title/authors/venue matched) | 2026-10-27 |
| 65 | E065 | Reputational herding: managers mimic others' decisions; shared evaluation produces correlated choices. | `scharfstein1990` | https://www.jstor.org/stable/2006678 (AER 80(3):465-479) | 1990 | IDEAS record (ideas.repec.org/a/aea/aecrev/v80y1990i3p465-79.html); no Crossref DOI | 2026-10-27 |
| 66 | E066 | Short-horizon speculators herd on the same information source, which can be unrelated to fundamentals. | `froot1992` | 10.1111/j.1540-6261.1992.tb04665.x | 1992 | Crossref (title/authors/venue matched) | 2026-10-27 |
| 67 | E067 | Algorithmic trading in FX: algorithmic strategies are more correlated with each other than human trades, yet improve price efficiency. | `chaboud2014` | 10.1111/jofi.12186 | 2014 | Crossref (title/authors/venue matched) | 2026-10-27 |
| 68 | E068 | HFTs trade in the direction of permanent price changes and against transitory pricing errors, improving price discovery. | `brogaard2014` | 10.1093/rfs/hhu032 | 2014 | Crossref (title/authors/venue matched) | 2026-10-27 |
| 69 | E069 | Review of the economics of HFT: speed, adverse selection, and liquidity consequences. | `menkveld2016` | 10.1146/annurev-financial-121415-033010 | 2016 | Crossref (title/authors/venue matched) | 2026-10-27 |
| 70 | E070 | The August 2007 quant unwind: common strategies produced simultaneous deleveraging, evidence of crowding among algorithmic funds. | `khandani2011` | 10.1016/j.finmar.2010.07.005 | 2011 | Crossref (title/authors/venue matched) | 2026-10-27 |
| 71 | E071 | Order-flow toxicity (VPIN) rose before the Flash Crash; liquidity provision withdrew when flow became toxic. | `easley2011` | 10.3905/jpm.2011.37.2.118 | 2011 | Crossref (title/authors/venue matched) | 2026-10-27 |
| 72 | E072 | News trading and speed: fast traders react to common signals, affecting price informativeness and liquidity. | `foucault2016` | 10.1111/jofi.12302 | 2016 | Crossref (title/authors/venue matched) | 2026-10-27 |
| 73 | E073 | Ultrafast extreme events (mini flash crashes) proliferate in a machine ecology beyond human response time. | `johnson2013` | 10.1038/srep02627 | 2013 | Crossref (title/authors/venue matched) | 2026-10-27 |

## (f) Stylized facts

| # | claim id | exact claim we will make in the paper | citation key | DOI or URL | year | how verified | verification date |
|---|---|---|---|---|---|---|---|
| 74 | F074 | Catalogue of stylized facts of asset returns (fat tails, absence of linear autocorrelation, volatility clustering, leverage effect) used as validation targets. | `cont2001` | 10.1080/713665670 | 2001 | Crossref (title/authors/venue matched) | 2026-10-27 |
| 75 | F075 | Speculative price changes have heavy (stable-Paretian) tails. | `mandelbrot1963` | 10.1086/294632 | 1963 | Crossref (title/authors/venue matched) | 2026-10-27 |
| 76 | F076 | ARCH: conditional heteroskedasticity captures volatility clustering. | `engle1982` | 10.2307/1912773 | 1982 | Crossref (title/authors/venue matched) | 2026-10-27 |
| 77 | F077 | GARCH generalization used to estimate volatility persistence in simulated returns. | `bollerslev1986` | 10.1016/0304-4076(86)90063-1 | 1986 | Crossref (title/authors/venue matched) | 2026-10-27 |
| 78 | F078 | Efficient markets hypothesis: definition of weak/semi-strong/strong form efficiency that our efficiency metrics operationalize. | `fama1970` | 10.2307/2325486 | 1970 | Crossref (title/authors/venue matched) | 2026-10-27 |
| 79 | F079 | Absolute returns display long memory; power-law decay of the autocorrelation of absolute returns. | `ding1993` | 10.1016/0927-5398(93)90006-D | 1993 | Crossref (title/authors/venue matched) | 2026-10-27 |
| 80 | F080 | Order-flow signs display long memory while prices remain nearly efficient: liquidity adapts to predictable flow. | `lillo2004` | 10.2202/1558-3708.1226 | 2004 | Crossref (title/authors/venue matched) | 2026-10-27 |

## (g) Contagion and spillovers

| # | claim id | exact claim we will make in the paper | citation key | DOI or URL | year | how verified | verification date |
|---|---|---|---|---|---|---|---|
| 81 | G081 | Spillover index from forecast-error variance decompositions measures return and volatility spillovers. | `diebold2009` | 10.1111/j.1468-0297.2008.02208.x | 2009 | Crossref (title/authors/venue matched) | 2026-10-27 |
| 82 | G082 | Generalized (order-invariant) directional spillover measures. | `diebold2012` | 10.1016/j.ijforecast.2011.02.006 | 2012 | Crossref (title/authors/venue matched) | 2026-10-27 |
| 83 | G083 | Variance decompositions define connectedness networks among financial entities; we apply them across assets and agent cohorts. | `diebold2014` | 10.1016/j.jeconom.2014.04.012 | 2014 | Crossref (title/authors/venue matched) | 2026-10-27 |
| 84 | G084 | Correlation-based contagion tests must adjust for heteroskedasticity; otherwise interdependence is mistaken for contagion. | `forbes2002` | 10.1111/0022-1082.00494 | 2002 | Crossref (title/authors/venue matched) | 2026-10-27 |
| 85 | G085 | Rational-expectations contagion: cross-market rebalancing by informed investors transmits shocks between assets with unrelated fundamentals. | `kodres2002` | 10.1111/1540-6261.00441 | 2002 | Crossref (title/authors/venue matched) | 2026-10-27 |
| 86 | G086 | Dense financial networks are robust to small shocks but fragile to large ones (phase transition). | `acemoglu2015` | 10.1257/aer.20130456 | 2015 | Crossref (title/authors/venue matched) | 2026-10-27 |
| 87 | G087 | Information heterogeneity among strategic speculators generates financial contagion; homogeneity reduces it. | `pasquariello2007` | 10.1093/rfs/hhl010 | 2007 | Crossref (title/authors/venue matched) | 2026-10-27 |

## (h) LLM agents in markets and economics; algorithmic collusion and monoculture

| # | claim id | exact claim we will make in the paper | citation key | DOI or URL | year | how verified | verification date |
|---|---|---|---|---|---|---|---|
| 88 | H088 | LLMs can be used as simulated economic agents (homo silicus) to replicate classic behavioural experiments at low cost. | `horton2023` | 10.3386/w31122 | 2023 | Crossref (title/authors/venue matched) | 2026-10-27 |
| 89 | H089 | EconAgent: LLM-empowered agents with perception, memory and reflection produce realistic macroeconomic dynamics in a simulation. | `li2024econagent` | 10.18653/v1/2024.acl-long.829 | 2024 | Crossref (title/authors/venue matched) | 2026-10-27 |
| 90 | H090 | ChatGPT-derived sentiment scores on headlines predict next-day stock returns out of sample. | `lopezlira2023chatgpt` | arXiv:2304.07619 (also SSRN 10.2139/ssrn.4412788) | 2023 | arXiv API + Crossref (SSRN) | 2026-10-27 |
| 91 | H091 | LLM economic forecasts are contaminated by memorization of historical data; backtests of LLM agents on real histories are unreliable. | `lopezlira2025memorization` | arXiv:2504.14765 | 2025 | arXiv API | 2026-10-27 |
| 92 | H092 | GPT choices in budgetary decision tasks score higher on rationality (GARP consistency) than humans and are largely stable across demographic prompts. | `chen2023pnas` | 10.1073/pnas.2316205120 | 2023 | Crossref (title/authors/venue matched) | 2026-10-27 |
| 93 | H093 | FinMem: single LLM trading agent with layered memory and character design evaluated by backtesting on real stock histories. | `yu2023finmem` | arXiv:2311.13743 | 2023 | arXiv API | 2026-10-27 |
| 94 | H094 | TradingGPT: multi-agent LLM trading system with layered memory and distinct characters, evaluated on historical returns. | `li2023tradinggpt` | arXiv:2309.03736 | 2023 | arXiv API | 2026-10-27 |
| 95 | H095 | FinAgent: multimodal foundation agent for financial trading, evaluated against trading benchmarks on historical data. | `zhang2024finagent` | 10.1145/3637528.3671801 | 2024 | Crossref (title/authors/venue matched) | 2026-10-27 |
| 96 | H096 | StockAgent: LLM-driven multi-agent stock-trading simulation with bulletin-board communication; GPT- vs Gemini-driven agents differ systematically in trading behaviour. | `zhang2024stockagent` | arXiv:2407.18957 | 2024 | arXiv API | 2026-10-27 |
| 97 | H097 | ASFM: agent-based simulated financial market with LLM traders and a real order-matching system; reactions to rate and inflation shocks match economic priors. | `gao2024asfm` | arXiv:2406.19966 | 2024 | arXiv API | 2026-10-27 |
| 98 | H098 | TwinMarket: scalable LLM-agent stock-market simulation with social-media interaction reproduces bubbles and crashes; reports that agents of one model family are more homogeneous. | `yang2025twinmarket` | arXiv:2502.01506 (NeurIPS 2025 proceedings page) | 2025 | arXiv API + proceedings.neurips.cc listing | 2026-10-27 |
| 99 | H099 | LLM agents in a persistent order-book simulation adhere to assigned strategies, generate price discovery and bubbles, and can be analysed with partial-dependence-style perturbations. | `lopezlira2025trade` | arXiv:2504.10789 | 2025 | arXiv API | 2026-10-27 |
| 100 | H100 | In Smith-Suchanek-Williams-type experimental markets, LLM agents price near fundamentals and form only muted bubbles unlike humans; LLM strategy text has lower variance. | `henning2025` | arXiv:2502.15800 | 2025 | arXiv API | 2026-10-27 |
| 101 | H101 | LLM agents in a simulated call auction exhibit disposition effects and extrapolative beliefs; prompt interventions causally change bubble size. | `ouyang2026` | arXiv:2604.18373 | 2026 | arXiv abstract page | 2026-10-27 |
| 102 | H102 | Across 15 LLMs, behaviour ranges from fundamental pricing to human-like bubbles; mixed populations are unstable and advanced models adapt to exploit weaker ones, raising volatility. | `saxena2026` | arXiv:2604.18602 | 2026 | arXiv abstract page | 2026-10-27 |
| 103 | H103 | Replications of Smith (1962) double auctions with LLM agents converge more slowly or not at all to competitive equilibrium, with model-family differences. | `struski2026` | arXiv:2609.02580 | 2026 | arXiv abstract page | 2026-10-27 |
| 104 | H104 | Market institution (call market, CDA, posted offer, bargaining) changes efficiency of identical LLM agents from 56% to 89%. | `chupilkin2026` | arXiv:2608.04020 | 2026 | arXiv abstract page | 2026-10-27 |
| 105 | H105 | FCLAgent: LLM-decided buy/sell with rule-based order sizing reproduces path-dependent price patterns absent from conventional ABMs. | `hashimoto2025` | arXiv:2510.12189 | 2025 | arXiv abstract page | 2026-10-27 |
| 106 | H106 | StockSim: open order-level simulator with latency, slippage and price-time matching for evaluating multi-agent LLMs. | `papadakis2025` | arXiv:2507.09255 | 2025 | arXiv abstract page | 2026-10-27 |
| 107 | H107 | LLM agents' trading-style switching in a year-long stock simulation is only partially consistent with behavioural-finance theory. | `li2026switch` | arXiv:2602.07023 | 2026 | arXiv API | 2026-10-27 |
| 108 | H108 | Overlap in retrieved information among RAG-based LLM trading agents raises mean pairwise action correlation from 0.04 to 0.41 and increases position volatility without raising profits. | `song2026` | https://openreview.net/forum?id=8v4FskZp8I | 2026 | ICLR 2026 workshop page (iclr.cc/virtual/2026/10016534); no arXiv version found | 2026-10-27 |
| 109 | H109 | LLM agents in multi-agent systems conform to peers; the confidence gap and format of peer information control the strength of herding. | `cho2025` | arXiv:2505.21588 | 2025 | arXiv abstract page | 2026-10-27 |
| 110 | H110 | LLM sellers in continuous double auctions collude more when they can communicate; collusion propensity differs across models. | `agrawal2025` | arXiv:2507.01413 | 2025 | arXiv abstract page | 2026-10-27 |
| 111 | H111 | LLM-based pricing agents autonomously reach supracompetitive prices; small prompt wording changes alter the degree of collusion. | `fish2024` | arXiv:2404.00806 | 2024 | arXiv API | 2026-10-27 |
| 112 | H112 | Q-learning pricing algorithms learn to collude tacitly in repeated oligopoly without communication. | `calvano2020` | 10.1257/aer.20190623 | 2020 | Crossref (title/authors/venue matched) | 2026-10-27 |
| 113 | H113 | Q-learning under sequential pricing converges to supracompetitive (Edgeworth-cycle) outcomes. | `klein2021` | 10.1111/1756-2171.12383 | 2021 | Crossref (title/authors/venue matched) | 2026-10-27 |
| 114 | H114 | Empirical evidence of higher margins after adoption of algorithmic pricing in German retail gasoline. | `assad2024` | 10.1086/726906 | 2024 | Crossref (title/authors/venue matched) | 2026-10-27 |
| 115 | H115 | Asymmetric pricing-algorithm adoption softens competition through commitment and frequency effects. | `brown2023` | 10.1257/mic.20210158 | 2023 | Crossref (title/authors/venue matched) | 2026-10-27 |
| 116 | H116 | Reinforcement-learning informed speculators in a Kyle-type market sustain supracompetitive profits via price-trigger strategies or homogenized learning biases, lowering price informativeness and liquidity. | `dou2025` | 10.3386/w34054 | 2025 | Crossref (title/authors/venue matched) | 2026-10-27 |
| 117 | H117 | Q-learning market makers learn adverse-selection pricing but quote markups over competitive spreads, shrinking with the number of algorithms. | `colliard2026` | 10.1093/rfs/hhag010 | 2026 | Crossref (title/authors/venue matched) | 2026-10-27 |
| 118 | H118 | Algorithmic monoculture: when many decision-makers rely on the same algorithm, outcomes can be worse even if the algorithm is individually more accurate. | `kleinberg2021` | 10.1073/pnas.2018340118 | 2021 | Crossref (title/authors/venue matched) | 2026-10-27 |
| 119 | H119 | Shared foundation models produce outcome homogenization: the same individuals are systematically mis-served across deployers. | `bommasani2022` | arXiv:2211.13972 | 2022 | arXiv API | 2026-10-27 |
| 120 | H120 | Foundation models create homogenization risk: downstream systems inherit the same strengths and failure modes. | `bommasani2021` | arXiv:2108.07258 | 2021 | arXiv API | 2026-10-27 |
| 121 | H121 | Generative AI raises individual creativity but reduces collective diversity of outputs. | `doshi2024` | 10.1126/sciadv.adn5290 | 2024 | Crossref (title/authors/venue matched) | 2026-10-27 |
| 122 | H122 | Standard LLM outputs show diminished diversity of thought relative to human samples. | `park2024brm` | 10.3758/s13428-023-02307-x | 2024 | Crossref (title/authors/venue matched) | 2026-10-27 |
| 123 | H123 | LLM assistance homogenizes human creative ideation across users. | `anderson2024` | 10.1145/3635636.3656204 | 2024 | Crossref (title/authors/venue matched) | 2026-10-27 |
| 124 | H124 | S3: social-network simulation with LLM-empowered agents reproduces emotion, attitude and information propagation. | `gao2023s3` | arXiv:2307.14984 | 2023 | arXiv API | 2026-10-27 |
| 125 | H125 | Generative agents with memory, reflection and planning produce believable emergent social behaviour. | `park2023` | 10.1145/3586183.3606763 | 2023 | Crossref (title/authors/venue matched) | 2026-10-27 |
| 126 | H126 | LLMs can simulate multiple humans and replicate classic human-subject studies (Turing experiments), with fidelity varying by model and task. | `aher2023` | https://proceedings.mlr.press/v202/aher23a.html | 2023 | publisher page (PMLR v202:337-371) + arXiv API 2208.10264 | 2026-10-27 |
| 127 | H127 | Conditioning LLMs on demographic backstories produces silicon samples that match human sub-populations' response distributions. | `argyle2023` | 10.1017/pan.2023.2 | 2023 | Crossref (title/authors/venue matched) | 2026-10-27 |
| 128 | H128 | Expectations elicited from an LLM reading WSJ articles track professional forecaster surveys and are extrapolative. | `bybee2023` | arXiv:2305.02823 | 2023 | arXiv API | 2026-10-27 |
| 129 | H129 | Automated social science: LLMs generate hypotheses and act as subjects in simulated experiments. | `manning2024` | arXiv:2404.11794 | 2024 | arXiv API | 2026-10-27 |
| 130 | H130 | ChatGPT choices in behavioural games fall within the human distribution and are more cooperative/altruistic than the median human. | `mei2024` | 10.1073/pnas.2313925121 | 2024 | Crossref (title/authors/venue matched) | 2026-10-27 |
| 131 | H131 | LLMs in repeated games cooperate well in self-interested games but coordinate poorly, revealing systematic behavioural signatures. | `akata2025` | 10.1038/s41562-025-02172-y | 2025 | Crossref (title/authors/venue matched) | 2026-10-27 |
| 132 | H132 | LLM service behaviour drifts over months (GPT-3.5/4 March vs June 2023), motivating frozen decision libraries for replicability. | `chen2024hdsr` | 10.1162/99608f92.5317da47 | 2024 | Crossref (title/authors/venue matched) | 2026-10-27 |
| 133 | H133 | Ensembles of twelve LLMs match human crowd forecasting accuracy (wisdom of the silicon crowd), illustrating diversity benefits across models. | `schoenegger2024` | 10.1126/sciadv.adp1528 | 2024 | Crossref (title/authors/venue matched) | 2026-10-27 |
| 134 | H134 | Multi-agent reinforcement-learning stock market reproduces stylized facts; RL-agent precursor of LLM-populated markets. | `lussange2021` | 10.1007/s10614-020-10038-w | 2021 | Crossref (title/authors/venue matched) | 2026-10-27 |
| 135 | H135 | Generative AI agents can be modelled as economic agents with new primitives (prompts, context); framework for LLM-agent economics. | `immorlica2024` | arXiv:2406.00477 | 2024 | arXiv API | 2026-10-27 |
| 136 | H136 | LLM social simulations are a promising method provided validity concerns (homogeneity, sycophancy, drift) are addressed. | `anthis2025` | arXiv:2504.02234 | 2025 | arXiv API | 2026-10-27 |
| 137 | H137 | Deep learning adoption increases uniformity and interconnectedness in finance; models built on the same data generate correlated predictions. | `gensler2020` | 10.2139/ssrn.3723132 | 2020 | Crossref (title/authors/venue matched) | 2026-10-27 |
| 138 | H138 | Widespread use of common AI models and data sources could increase correlations in trading, lending and pricing (FSB 2024). | `fsb2024` | https://www.fsb.org/uploads/P14112024.pdf | 2024 | publisher page (fsb.org, 14 Nov 2024) | 2026-10-27 |
| 139 | H139 | AI in risk management can create endogenous systemic risk through common models and procyclical responses. | `danielsson2022` | 10.1016/j.jbankfin.2021.106290 | 2022 | Crossref (title/authors/venue matched) | 2026-10-27 |
| 140 | H140 | AI amplifies existing vulnerabilities and makes crises faster and more severe in a game-theoretic model of AI-driven markets. | `danielsson2025` | 10.1016/j.jfs.2025.101453 | 2025 | Crossref (title/authors/venue matched) | 2026-10-27 |
| 141 | H141 | Intelligent financial system: AI transforms intermediation, asset management and payments, with monoculture and correlation among the financial-stability concerns. | `aldasoro2025` | 10.1016/j.jfs.2025.101472 | 2025 | Crossref (title/authors/venue matched) | 2026-10-27 |
| 142 | H142 | In LLM re-creations of herding experiments, AI agents rely more on private signals than humans, but can be prompted to herd when maximizing profit. | `hansen2025` | 10.17016/FEDS.2025.090 | 2025 | Crossref (title/authors/venue matched) | 2026-10-27 |
| 143 | H143 | Unified model in which AI adoption share and signal correlation raise systemic-risk coupling convexly; 13F evidence of tail-loss amplification. | `meng2026` | arXiv:2604.03272 | 2026 | arXiv abstract page | 2026-10-27 |

## Summary counts

- Verified entries in this ledger: **143**, one per bib key; `paper/references.bib` has 143 entries (every ledger row has a bib entry and vice versa; cross-checked with `comm` on 2026-10-27).
- Could not verify / replaced: see next section (none of these are in the bib).

## Could not verify (NOT in references.bib)

| item | what we looked for | why excluded |
|---|---|---|
| "ElliottAgents" / other 2025 LLM-trading benchmark names seen in secondary blogs | a primary arXiv or proceedings record | no primary record found; blog mentions only |
| Danielsson & Uthemann, "Artificial intelligence and financial crises" arXiv:2407.17048 | journal reference on the arXiv page | arXiv page shows no journal-ref; we cite the Journal of Financial Stability version `danielsson2025` (Crossref 10.1016/j.jfs.2025.101453) instead |
| Lopez-Lira & Tang (2023) journal version | a peer-reviewed journal DOI | still an arXiv/SSRN preprint as of v6 (Oct 2025); cited as `lopezlira2023chatgpt` (arXiv) |
| Dou, Goldstein & Ji — journal version | a journal DOI | only NBER w34054 / SSRN 4452704 exist; cited as `dou2025` (NBER, Crossref 10.3386/w34054) |
| Colliard, Foucault & Lovo — volume/pages | RFS volume, issue, pages | Crossref lists the RFS article (10.1093/rfs/hhag010) with no volume/pages yet; bib entry carries `note = {Forthcoming}` |
| Song (2026) — arXiv version | an arXiv ID for the ICLR 2026 workshop paper | none found; cited via the OpenReview/ICLR page |
| Byrd, Hybinette & Balch 2019 arXiv 1904.12066 | arXiv API record (rate-limited, HTTP 429) | abstract page fetched instead; we cite the SIGSIM-PADS 2020 version `byrd2020` (Crossref 10.1145/3384441.3395986) |
| Arthur, Holland, LeBaron, Palmer & Tayler (1997) Santa Fe chapter | a DOI for the book chapter in *The Economy as an Evolving Complex System II* | no Crossref record; we cite `lebaron1999` (JEDC) for the Santa Fe market instead |
| Hasbrouck (2007) *Empirical Market Microstructure* (book) | a DOI | book has no DOI; not needed since `hasbrouck1991`/`hasbrouck1993`/`hasbrouck1995` cover the measures |
| Vives (2008) *Information and Learning in Markets* | a DOI | book without Crossref DOI; `hellwig1980`, `grossman1980`, `kyle1985` suffice |
| Aher et al. 2023 ICML via ACM DL (10.5555/3618408.3618425) | Crossref record | 10.5555 is not registered at Crossref; cited via the PMLR page instead |
| Aldasoro et al. — guessed DOI 10.1016/j.jfs.2025.101456 | Crossref | resolved to a different paper; corrected to 10.1016/j.jfs.2025.101472 (now `aldasoro2025`) |
| Thurner, Farmer & Geanakoplos — guessed DOI 10.1080/14697688.2011.586336 | Crossref | not found; corrected to 10.1080/14697688.2012.674301 (now `thurner2012`) |
| Pasquariello — guessed DOI 10.1093/rfs/hhl046 | Crossref | not found; corrected to 10.1093/rfs/hhl010 (now `pasquariello2007`) |
| Chiarella–Iori–Perelló, Back–Cao–Willard, Foster–Viswanathan, Ladley, Hasbrouck 1991, Kondor 2012 — first-guess DOIs | Crossref | each first guess resolved to a *different* paper; all corrected by Crossref bibliographic query and re-verified (DOIs in table above) |