SPB Git

Repository Public

TimeSeriesTests.swift 9.3 KB 135 lines swift

1import Foundation2import Testing3import SheetCore4import SheetUI5import Statistics6import AnalysisKit7@testable import StatsModels89/// Monthly series with trend and seasonality, first-of-month dates.10func seasonalSeries(n: Int = 72, seed: UInt64 = 4) -> (dates: [Double], y: [Double]) {11    var rng = Xoshiro256StarStar(seed: seed)12    let season = [-8.0, -5, 0, 3, 6, 9, 11, 8, 3, -2, -9, -16]13    let y = (0..<n).map { 100 + 0.8 * Double($0) + season[$0 % 12] + 2 * rng.nextNormal() }14    let dates = (0..<n).map { DateSerial.serial(year: 2020 + $0 / 12, month: 1 + $0 % 12, day: 1) }15    return (dates, y)16}1718@MainActor @Suite struct TimeSeriesTests {19    let data = seasonalSeries()20    var s: WorkbookSession { session(["date", "y"], data.y.indices.map { [data.dates[$0], data.y[$0]] }) }2122    @Test func acfAndPacfMatchTheEngine() throws {23        let r = try run(AutocorrelationTool.descriptor, s, range: "A1:B73") { $0.set("series", column: 1); $0.set("maxLag", int: 12) }24        let acf = try TimeSeries.acf(data.y, maxLag: 12), pacf = try TimeSeries.pacf(data.y, maxLag: 12)25        let t = try #require(r.table(titled: "Autocorrelations"))26        #expect(t.rows.count == 13)27        #expect(close(t.number(1, 1), acf.values[1]) && close(t.number(12, 1), acf.values[12]) && close(t.number(3, 5), pacf.values[3]))28        #expect(close(t.number(2, 3), acf.lower[2]) && close(t.number(2, 4), acf.upper[2]))29        #expect(r.charts.count == 2)30        // differenced first31        let d = try run(AutocorrelationTool.descriptor, s, range: "A1:B73") { $0.set("series", column: 1); $0.set("maxLag", int: 6); $0.set("diff", int: 1) }32        #expect(close(try #require(d.table(titled: "Autocorrelations")).number(1, 1), try TimeSeries.acf(TimeSeries.difference(data.y), maxLag: 6).values[1]))33    }3435    @Test func differencingWritesAlignedValuesAndUndoes() throws {36        let session = s37        let r = try run(DifferencingTool.descriptor, session, range: "A1:B73") { $0.set("series", column: 1); $0.set("lag", int: 12) }38        let ref = TimeSeries.difference(data.y, lag: 12)39        let t = try #require(r.table(titled: "Before and after"))40        #expect(t.number(2, 1) == 60 && close(t.number(2, 2), Descriptive.mean(ref)))41        let before = snapshot(session.workbook)42        _ = try #require(r.actions.first).perform(session)43        let sh = session.workbook.sheets[0]44        #expect(sh.value(12, 2).isBlank)                                        // first 12 observations have no difference45        #expect(close(sh.value(13, 2).numberValue ?? .nan, ref[0]) && close(sh.value(72, 2).numberValue ?? .nan, ref[59]))46        session.undo()47        #expect(snapshot(session.workbook) == before)48        #expect(throws: ToolError.self) { try run(DifferencingTool.descriptor, session, range: "A1:B73") { $0.set("series", column: 1); $0.set("lag", int: 40); $0.set("order", int: 2) } }49    }5051    @Test func holtWintersForecastsMatchTheEngineAndAreWrittenBelowTheSeries() throws {52        let session = s53        let r = try run(ExponentialSmoothingTool.descriptor, session, range: "A1:B73") { $0.set("series", column: 1); $0.set("time", column: 0); $0.set("horizon", int: 6) }54        let fit = try ExponentialSmoothing.fit(data.y, trend: .additive, seasonal: .additive, period: 12)55        let interval = try #require(fit.forecastInterval(6))56        let f = try #require(r.table(titled: "Forecasts"))57        #expect(f.rows.count == 7)58        #expect(f.number(1, 1) == DateSerial.serial(year: 2026, month: 1, day: 1) && f.number(6, 1) == DateSerial.serial(year: 2026, month: 6, day: 1))59        #expect(close(f.number(1, 2), fit.forecast(6)[0]) && close(f.number(6, 4), interval.lower[5]) && close(f.number(6, 5), interval.upper[5]))60        #expect(close(try #require(r.table(titled: "Parameters")).value("Level smoothing α"), fit.parameters.alpha))61        #expect(close(try #require(r.table(titled: "Accuracy")).value("Root mean squared error"), fit.rmse))62        let before = snapshot(session.workbook)63        _ = try #require(r.actions.first { $0.id == "write-forecasts" }).perform(session)64        #expect(close(session.workbook.sheets[0].value(73, 1).numberValue ?? .nan, fit.forecast(6)[0]))65        #expect(close(session.workbook.sheets[0].value(78, 1).numberValue ?? .nan, fit.forecast(6)[5]))66        session.undo()67        #expect(snapshot(session.workbook) == before)68    }6970    @Test func multiplicativeSeasonalityHasNoIntervalAndSimpleModelsWork() throws {71        let m = try run(ExponentialSmoothingTool.descriptor, s, range: "A1:B73") { $0.set("series", column: 1); $0.set("model", choice: "hwMultiplicative"); $0.set("horizon", int: 4) }72        let f = try #require(m.table(titled: "Forecasts"))73        #expect(f.columnCount == 2)                                             // horizon + forecast only74        let ref = try ExponentialSmoothing.fit(data.y, trend: .additive, seasonal: .multiplicative, period: 12)75        #expect(close(f.number(4, 1), ref.forecast(4)[3]))76        let simple = try run(ExponentialSmoothingTool.descriptor, s, range: "A1:B73") { $0.set("series", column: 1); $0.set("model", choice: "simple"); $0.set("horizon", int: 3) }77        #expect(close(try #require(simple.table(titled: "Forecasts")).number(1, 1), try ExponentialSmoothing.fit(data.y).forecast(1)[0]))78        // too short for a seasonal model79        let short = session(["y"], (0..<15).map { [Double($0)] })80        #expect(throws: ToolError.self) { try run(ExponentialSmoothingTool.descriptor, short, range: "A1:A16") { $0.set("series", column: 0) } }81    }8283    @Test func arimaMatchesTheEngine() throws {84        let r = try run(ARIMATool.descriptor, s, range: "A1:B73") { $0.set("series", column: 1); $0.set("p", int: 1); $0.set("d", int: 1); $0.set("q", int: 0); $0.set("horizon", int: 5) }85        let fit = try ARIMA.fit(data.y, order: .init(p: 1, d: 1, q: 0))86        let fc = try fit.forecast(5)87        let p = try #require(r.table(titled: "Parameters"))88        #expect(p.rows.count == fit.parameterNames.count + 1)89        #expect(close(p.number(1, 1), fit.estimates[0]) && close(p.number(1, 2), fit.standardErrors[0] ?? .nan))90        #expect(p.rows.last?[3].isBlank == true)                                // no z test for σ²91        let f = try #require(r.table(titled: "Forecasts"))92        #expect(close(f.number(5, 1), fc.mean[4]) && close(f.number(5, 3), fc.lower[4]) && close(f.number(5, 4), fc.upper[4]))93        #expect(close(r.metric("AIC") ?? .nan, fit.aic))94        let lb = try #require(r.table(titled: "Ljung–Box test on the residuals"))95        #expect(close(lb.value("Q statistic"), try RegressionDiagnostics.ljungBox(fit.residuals, lags: 10, modelDF: 1).statistic))96        #expect(r.title == "ARIMA(1, 1, 0)")97    }9899    @Test func garchAndADFMatchTheEngine() throws {100        var rng = Xoshiro256StarStar(seed: 8)101        var variance = 1e-4, shock = 0.0102        let returns: [Double] = (0..<400).map { _ in103            variance = 4e-6 + 0.1 * shock * shock + 0.85 * variance104            shock = variance.squareRoot() * rng.nextNormal()105            return 0.0005 + shock106        }107        let rs = session(["r"], returns.map { [$0] })108        let g = try run(GARCHTool.descriptor, rs, range: "A1:A401") { $0.set("series", column: 0) }109        let fit = try GARCH.fit(returns.map { $0 * 100 })110        let est = try #require(g.table(titled: "Estimates"))111        #expect(close(est.value("ω"), fit.parameters.omega) && close(est.value("Persistence α + β"), fit.persistence))112        #expect(close(try #require(g.table(titled: "Variance forecasts")).number(3, 1), fit.forecastVariance(10)[2]))113        _ = try #require(g.actions.first).perform(rs)114        #expect(close(rs.workbook.sheets[0].value(400, 1).numberValue ?? .nan, fit.conditionalVolatility[399]))115116        let a = try run(ADFTool.descriptor, s, range: "A1:B73") { $0.set("series", column: 1); $0.set("regression", choice: "ct"); $0.set("lagRule", choice: "fixed"); $0.set("lags", int: 2) }117        let ref = try UnitRootTest.augmentedDickeyFuller(data.y, regression: .constantTrend, lags: .fixed(2))118        let t = try #require(a.table(titled: "Augmented Dickey–Fuller test"))119        #expect(close(t.value("ADF statistic (τ)"), ref.statistic) && close(t.value("p-value (MacKinnon, lower tail)"), ref.pValue))120        #expect(t.value("Lags used") == 2 && close(t.value("Critical value 5 %"), ref.critical5))121        let sentence = a.blocks.compactMap { b -> String? in if case .text(let t) = b, t.kind == .interpretation { t.lines.first } else { nil } }.first ?? ""122        #expect(sentence.contains(ref.pValue < 0.05 ? "reject the null hypothesis" : "fail to reject the null hypothesis") && sentence.contains("unit root"))123        let auto = try run(ADFTool.descriptor, s, range: "A1:B73") { $0.set("series", column: 1) }124        #expect(close(auto.metric("ADF statistic (τ)") ?? .nan, try UnitRootTest.augmentedDickeyFuller(data.y).statistic))125    }126127    @Test func missingObservationsAreSkippedAndReported() throws {128        var rows: [[Any?]] = data.y.indices.map { [data.dates[$0], data.y[$0]] }129        rows[10][1] = nil130        let r = try run(ADFTool.descriptor, session(["date", "y"], rows), range: "A1:B73") { $0.set("series", column: 1) }131        let notes = r.blocks.compactMap { b -> [String]? in if case .text(let t) = b, t.kind == .note { t.lines } else { nil } }.flatMap { $0 }132        #expect(notes.contains { $0.contains("1 rows without a number") })133    }134}135
Files main
Menu

Theme