Research methodology
Market research through financial econometrics, deep learning and behavioural simulation. Understand the measurement, its interpretation and the conditions under which it applies.
Explore 60 analysis methodsOne question, several research layers
Center Quant combines financial econometrics, risk analysis, time-series deep learning and behavioural simulation. Each addresses a part of the research question: describing market conditions, estimating a continuation or exploring responses to changing conditions.
60 analysis methods
The analysis library has five groups: technical and price structure; news, events and sentiment; fundamentals, macro and flows; statistics; strategies and validation. Quantitative calculations measure data properties. Information-research workflows add economic context, while strategy evaluation examines evidence from retained results.
Method library
Choose a research group, then open a method for its principle and interpretation.
Technical & price structure
9 methods
Indicators & trendMoving averages · RSI · MACD
Measure trend and momentum in a price sequence.
Principle
Moving averages describe the price baseline; RSI compares smoothed gains and losses; MACD compares two exponential averages.
MAₙ(t) = (Pₜ + … + Pₜ₋ₙ₊₁) / n
EMAₜ = αPₜ + (1 − α)EMAₜ₋₁; α = 2 / (n + 1)
RSI = 100 − 100 / (1 + G / L)
MACDₜ = EMA_fast(P)ₜ − EMA_slow(P)ₜ
P is the closing price; G and L are average gains and losses under the indicator’s smoothing convention.
- Interpretation
- Read slope, price position and agreement between indicators in the context of trend.
- Assumptions and scope
- Indicators lag prices. A threshold or crossover needs evaluation before use as a trading rule.
Related reading: Murphy (1999)
Price structureSwing structure · Support and resistance
Locate price zones and the structure of market swings.
Principle
Price pivots identify support, resistance and the structure of market swings.
- Interpretation
- Price relative to these zones helps frame continuation or structural-break conditions.
- Assumptions and scope
- These zones describe observed prices; they do not guarantee a future price reaction.
Related reading: Murphy (1999)
Candlestick patternsOHLC pattern recognition
Recognise candle and multi-candle formations.
Principle
Compare bodies, wicks, ranges and relationships between candles using pattern rules.
- Interpretation
- Patterns add information about price responses in the area under study.
- Assumptions and scope
- A pattern name does not establish reversal probability; trend context and completed observations are required.
Related reading: Murphy (1999)
IchimokuIchimoku Kinkō Hyō
Read trend and balance through several reference lines.
Principle
High–low range midpoints form Tenkan, Kijun and the cloud. Signals combine crossovers with price position relative to the cloud.
- Interpretation
- Price above, within or below the cloud describes trend conditions and transition zones.
- Assumptions and scope
- Cloud and Chikou plot offsets are drawing positions; they do not turn current data into future observations.
Related reading: Murphy (1999)
Elliott WaveElliott Wave · Swing sequences
Study wave structure through price pivots.
Principle
Pivot sequences are compared with impulse, corrective-wave and ratio rules.
- Interpretation
- Matching counts suggest a price structure to assess against subsequent observations.
- Assumptions and scope
- A sequence can support several counts. Pivots require confirmation time, and developing patterns can change.
Related reading: Murphy (1999)
HarmonicHarmonic patterns · Fibonacci geometry
Study price structures through ratios between swings.
Principle
Successive pivots are compared with the geometric ratios of harmonic formations.
- Interpretation
- A matching formation highlights a price zone to observe; it does not confirm reversal.
- Assumptions and scope
- Ratio tolerances and pivot selection affect results; a completion zone does not confirm reversal.
Related reading: Murphy (1999)
SMCMarket structure · BOS · CHoCH · Fair value gaps
Describe structural changes and price gaps.
Principle
Swings and break prices identify BOS and CHoCH; candle-range relationships identify FVG zones.
- Interpretation
- Structures are observations at analysis time for forming subsequent hypotheses.
- Assumptions and scope
- Patterns may revise as data arrives. SMC labels do not establish institutional orders or historical signal availability.
Related reading: Murphy (1999)
ChanlunChan theory · Fractals · Strokes and segments
Decompose price structure into nested swing levels.
Principle
Rules for candle inclusion, fractals and strokes form segments and central zones.
- Interpretation
- Relationships between swings and central zones describe structural conditions.
- Assumptions and scope
- Availability depends on the specialist calculator, interval and volume data. Unconfirmed segments may change.
Intraday analysisIntraday trend and momentum
Read short-horizon intraday conditions.
Principle
Apply price indicators to intraday observations to compare trend, momentum and range.
- Interpretation
- Signals are interpreted against the selected interval and trading-session conditions.
- Assumptions and scope
- Noise, session gaps and trading costs have greater influence at short intervals.
Related reading: Murphy (1999)
News, events & sentiment
8 methods
News fundamentalsEvidence-based financial news analysis
Understand the economic content of published information.
Principle
Read the material, identify timing and affected participants, then compare accounts and market responses.
- Interpretation
- Distinguish confirmed events, interpretation and impact hypotheses.
- Assumptions and scope
- Conclusions stay within the material examined; repeated stories or unclear timing weaken the evidence.
Event-driven researchEvent-driven transmission analysis
Explore an event’s transmission channels.
Principle
Identify the shock, affected groups, transmission mechanisms and conditions that can alter the response.
- Interpretation
- Read the impact chain and observable developments that could test the hypothesis.
- Assumptions and scope
- Qualitative analysis requires dated evidence; simultaneous developments do not establish causation.
Related reading: MacKinlay (1997)
Event studyMean-adjusted event study · Abnormal returns
Measure price responses around timestamped events.
Principle
Returns around an event are compared with a preceding estimation-window mean. Differences are accumulated over the event window.
ARₜ = rₜ − μ̂_pre
CAR[a,b] = ∑ₜ₌ₐᵇ ARₜ
r is the simple return; μ̂_pre is the mean return in the pre-event estimation window.
- Interpretation
- AR and CAR describe deviations from the estimated return baseline.
- Assumptions and scope
- Complete pre- and post-event windows are required. Daily returns can include pre-announcement hours, and events can overlap.
Related reading: MacKinlay (1997)
Economic calendarMacroeconomic announcement framework
Track announcements that may change market expectations.
Principle
Compare release schedules, previous values, expectations and actuals with policy context when these inputs are available.
- Interpretation
- Differences between new information and expectations help frame response hypotheses.
- Assumptions and scope
- Schedules may change; revised figures and absent consensus forecasts require separate treatment.
Related reading: MacKinlay (1997)
Market sentimentQualitative sentiment assessment
Assess expressed information tone and expectations.
Principle
Read relevant material, classify viewpoints and compare them with price context.
- Interpretation
- Results describe sentiment within the material examined.
- Assumptions and scope
- The information sample may be biased. A sentiment assessment is not a market-wide index or trading probability.
Related reading: Kahneman (2002)
Behavioral financeBehavioural finance · Reference dependence
Examine how biases can affect decisions.
Principle
Compare observed behaviour with mechanisms such as loss aversion, anchoring and herding.
- Interpretation
- These mechanisms support explanations and testable hypotheses.
- Assumptions and scope
- Individual motives cannot be inferred from price or a single story; relevant behavioural evidence is needed.
Related reading: Kahneman (2002)
Geopolitics & regulationGeopolitical and regulatory transmission
Analyse policy effects and geopolitical risk.
Principle
Examine changes in rules, trade, supply chains and capital flows alongside the event’s applicable scope.
- Interpretation
- Interpret effects by asset group and the conditions that could change developments.
- Assumptions and scope
- Distinguish enacted measures, proposals and statements; market responses also reflect other influences.
Fundamentals, macro & flows
13 methods
Macro & economic cyclesMacroeconomic cycle analysis
Place assets in the economic-cycle context.
Principle
Compare growth, inflation, interest rates and financial conditions against release timing.
- Interpretation
- Read changes across real activity, policy and valuation expectations.
- Assumptions and scope
- Economic data are delayed and may be revised; relationships can shift across policy regimes.
Forex fundamentalsRelative macro and monetary-policy analysis
Study the relative drivers of a currency pair.
Principle
Compare rates, policy outlook, inflation and economic activity across the two economies.
- Interpretation
- Differentials and changing expectations help explain exchange-rate pressures.
- Assumptions and scope
- Relevant data for both sides are required; a rate differential alone does not establish an exchange-rate forecast.
Commodity supply & demandCommodity supply, demand and inventories
Understand physical commodity-market drivers.
Principle
Examine production, consumption, inventories and supply disruptions across cycles and seasons.
- Interpretation
- Changes in supply-demand balance provide context for price analysis.
- Assumptions and scope
- Fundamental coverage varies by commodity. A price series alone does not supply physical supply-demand data.
Precious metalsPrecious-metal monetary and physical drivers
Study monetary and demand drivers of precious metals.
Principle
Compare real rates, quote-currency conditions and available demand, reserve or investment evidence.
- Interpretation
- Distinguish monetary effects from each metal’s physical drivers.
- Assumptions and scope
- Specialist evidence may cover gold only. Gold evidence does not automatically represent silver, platinum or palladium.
Market positioning (COT)Commitments of Traders · Positioning analysis
Examine positioning in published reports.
Principle
Compare long, short and net positions of participant groups across matching report periods.
- Interpretation
- Net positions and their changes describe concentration among reported participants.
- Assumptions and scope
- Reports are delayed and have a defined scope. Reference-contract positioning is not total spot-market positioning.
Crypto project fundamentalsProject fundamentals and token economics
Evaluate a crypto project’s economic model.
Principle
Examine the product, token mechanics, governance and verifiable activity measures.
- Interpretation
- Compare usage and token-holder rights with the research thesis.
- Assumptions and scope
- Project claims need independent corroboration; network activity does not necessarily accrue revenue to the token.
Token unlock & treasuryToken supply schedules and treasury analysis
Study changes in circulating supply and treasury holdings.
Principle
Compare unlock schedules, recipients and treasury information against published records.
- Interpretation
- Distribution size and timing identify developments worth monitoring.
- Assumptions and scope
- Unlocking does not imply selling. Schedules can change and treasury-wallet identification may be incomplete.
On-chain researchOn-chain activity and entity analysis
Examine recorded blockchain activity.
Principle
Compare transactions, addresses and transfers under the definitions of the examined data.
- Interpretation
- Results describe activity within the covered network and available entity labels.
- Assumptions and scope
- Addresses are not users. Internal transfers and incomplete wallet labels can distort flow interpretation.
Stablecoin flowsStablecoin supply and transfer analysis
Explore liquidity associated with stablecoins.
Principle
Examine supply changes, minting, burning and transfers by network and verified destination.
- Interpretation
- Transfers and supply changes add liquidity context.
- Assumptions and scope
- Bridges and internal transfers can be double-counted; minting does not establish that capital entered trading markets.
DeFi yieldsDeFi yield decomposition
Understand yield sources and protocol risks.
Principle
Separate lending income, fees, token incentives and reinvestment mechanics using disclosed information.
- Interpretation
- Read yield alongside principal assets, costs and sustainability conditions.
- Assumptions and scope
- APR and APY differ; displayed yields can change and may omit contract, liquidity or depegging risks.
ETF flowsETF subscriptions, redemptions and asset flows
Track reported fund flows.
Principle
Compare inflows, outflows and asset changes against reporting periods and fund structure.
- Interpretation
- Flows describe activity within the observed fund sample.
- Assumptions and scope
- Asset growth can reflect price changes. Secondary-market trading volume is not net fund flow.
Investor lensesComparative investment frameworks
Compare an asset through several investment frameworks.
Principle
Examine assumptions about value, growth, quality or momentum and the evidence supporting each thesis.
- Interpretation
- Differences reveal which assumptions drive the conclusions.
- Assumptions and scope
- A lens is a reasoning framework; valuation cannot be completed without suitable financial inputs.
Funding & derivatives positioningFunding and derivatives positioning context
Read funding and positioning context in suitable retained results.
Principle
Compare funding, open interest and related observations when verified data are available.
- Interpretation
- Measures describe the observed contract market’s structure.
- Assumptions and scope
- This entry is retained for historical material; new derivatives research is currently unavailable in Center Quant.
Statistics, timeframes & probabilities
18 methods
Return distributionSimple returns · Sample moments
Describe observed returns, dispersion and asymmetry.
Principle
Returns are proportional changes in closing price. The mean, sample standard deviation and quantiles summarise their historical distribution.
rₜ = Pₜ / Pₜ₋₁ − 1
r̄ = Σᵢ rᵢ / N
s² = Σᵢ(rᵢ − r̄)² / (N − 1)
N is the number of return observations; r̄ is the sample mean; s is the sample standard deviation.
- Interpretation
- Quantiles locate outcomes within the sample; standard deviation describes dispersion around the mean.
- Assumptions and scope
- Historical sample properties can change. Return calculations require positive prices and consistent units.
Related reading: Campbell, Lo & MacKinlay (1997)
Stationarity (ADF)Augmented Dickey–Fuller (ADF)
Examine evidence of a unit root before interpreting relationships between price series.
Principle
ADF regresses changes on the lagged level and past changes. Its null hypothesis is a unit root; lag selection uses the AIC information criterion.
Δxₜ = a + ρxₜ₋₁ + Σᵢ γᵢΔxₜ₋ᵢ + εₜ
Δx is the change between observations; a is a constant; ε is the residual. The test examines ρ = 0.
- Interpretation
- A p-value below the significance level supports rejecting the unit-root null for the tested specification.
- Assumptions and scope
- Results depend on the window, lag choice and structural breaks. Failure to reject does not prove a unit root.
Related reading: Dickey & Fuller (1979)
Volatility forecast (GARCH)GARCH(1,1)
Estimate volatility and the persistence of high-volatility periods.
Principle
Current variance depends on the previous shock and previous variance. The applied variant uses a constant mean and normally distributed innovations.
rₜ = μ + εₜ
σ²ₜ = ω + αε²ₜ₋₁ + βσ²ₜ₋₁
r is the return; ε is the innovation; σ² is conditional variance. α + β measures volatility persistence.
- Interpretation
- The forecast estimates fluctuation size and the risk environment. A finite long-run variance exists when α + β is below one.
- Assumptions and scope
- The fit must converge and the data must suit the specification. A volatility forecast does not determine price direction.
Related reading: Bollerslev (1986)
Volatility regimeMarkov-switching volatility regimes
Identify lower- and higher-volatility states.
Principle
A Markov-switching model estimates return regimes with different variances and transition probabilities.
- Interpretation
- Current state membership describes compatibility with the fitted regimes.
- Assumptions and scope
- The model must converge. Volatility-state membership is not the probability of a price rise or fall.
Related reading: Campbell, Lo & MacKinlay (1997)
Asset correlationPearson correlation
Measure linear co-movement between two assets in the same research window.
Principle
Correlation normalises return covariance by each series’ standard deviation.
ρₓᵧ = Cov(rₓ, rᵧ) / (σₓσᵧ)
Cov is covariance; σ is standard deviation; both series share observation timestamps.
- Interpretation
- Values near 1 or −1 indicate strong positive or negative linear co-movement in the observed sample.
- Assumptions and scope
- Correlation can change with market conditions and does not establish causation.
Related reading: Campbell, Lo & MacKinlay (1997)
Cointegration & spreadEngle–Granger cointegration · OLS spread
Examine whether a combination of two price series is more stable than either series alone.
Principle
Engle–Granger tests the residual of a relationship between two series for stationarity. Ordinary least squares with an intercept estimates the coefficient and spread.
yₜ = a + βxₜ + εₜ
sₜ = yₜ − a − βxₜ
β is the regression coefficient; a is the intercept; s is the spread. The test null is no cointegration.
- Interpretation
- Evidence of cointegration supports studying the spread and departures from its historical relationship.
- Assumptions and scope
- Series must share timestamps and meet test assumptions. Historical relationships can break; a regression coefficient alone does not define an executable trade.
Related reading: Engle & Granger (1987)
Mean reversion (OU)Ornstein–Uhlenbeck process
Estimate a series’ tendency to return toward equilibrium.
Principle
After the stationarity check passes, an OU fit estimates the equilibrium level, reversion speed and noise.
- Interpretation
- Reversion speed and half-life describe the persistence of deviations in the sample.
- Assumptions and scope
- Reversion may disappear after a structural change; passing a test does not establish strategy profitability.
Related reading: Dickey & Fuller (1979) · Campbell, Lo & MacKinlay (1997)
Lead–lag relationshipsGranger predictive-association test
Explore predictive lead–lag relationships between two series.
Principle
Test whether lagged secondary price levels improve explanation of primary prices beyond the primary’s own lags.
- Interpretation
- Read the p-value against the null of no predictive contribution from the secondary series.
- Assumptions and scope
- The current calculation uses price levels without an automatic stationarity check at this step; predictive association does not establish causality.
Related reading: Campbell, Lo & MacKinlay (1997)
Independent bootstrapIndependent Percentile Bootstrap
Assess uncertainty in historical mean returns and return-to-volatility ratios.
Principle
Resample observations independently and calculate the statistic on each sample. Result percentiles form a confidence interval.
- Observed returns
- Independent resampling
- Sample statistics
- Confidence interval
- Interpretation
- The interval shows sensitivity to historical sampling; it is not a future price range.
- Assumptions and scope
- Independent resampling does not preserve serial dependence. Interpret the result under that assumption.
Related reading: Künsch (1989)
Tail risk (VaR/CVaR)Historical Value at Risk (VaR) · Conditional VaR (CVaR)
Assess a loss threshold and average loss in the adverse distribution tail.
Principle
Historical VaR uses an ordered return observation without interpolation. CVaR averages all returns at or below that threshold and reverses the sign to express loss.
k = max(1, ⌈(1 − c)N⌉)
VaR(c) = −r₍ₖ₎
CVaR(c) = −mean{rᵢ : rᵢ ≤ r₍ₖ₎}
c is the confidence level; r₍ₖ₎ is the kth smallest return. These formulas describe one observation period; all threshold ties enter CVaR.
- Interpretation
- VaR identifies a tail threshold; CVaR describes outcomes inside that tail. A sample with no tail losses can produce a negative signed loss.
- Assumptions and scope
- These estimates depend on historical samples and can miss unseen shocks. The confidence level is not the probability of a profitable trade.
Related reading: Rockafellar & Uryasev (2000)
Extreme tail risk (EVT)Extreme Value Theory · Peaks over threshold
Study the extreme-loss part of a distribution.
Principle
Loss-threshold exceedances are fitted with a generalised Pareto distribution to describe tail shape.
- Interpretation
- The shape parameter and its estimation error help distinguish heavy, approximately exponential and bounded tails.
- Assumptions and scope
- Results depend on the threshold and tail sample size. The current in-sample fit is not out-of-sample calibrated.
Related reading: Embrechts, Klüppelberg & Mikosch (1997)
Tail dependenceCopula dependence · Kendall’s tau
Explore dependence beyond linear correlation.
Principle
Kendall’s tau of two return series is mapped to a copula parameter under the selected family.
- Interpretation
- Tail-dependence coefficients describe joint extremes under the assumed copula family.
- Assumptions and scope
- Aligned observations and valid parameters are required. Dependence coefficients are model results, not proof of calibrated real-world risk.
Related reading: Nelsen (2006)
Drawdown & recoveryDrawdown · Maximum Drawdown (MDD)
Measure decline depth and observed recovery time.
Principle
Each observation is compared with the running peak of the same series. Maximum drawdown is the largest resulting decline.
Hₜ = max(P₁, …, Pₜ)
DDₜ = 1 − Pₜ / Hₜ
MDD = maxₜ DDₜ
H is the running peak; DD expresses a loss as a positive value. For an equity curve, equity replaces P.
- Interpretation
- MDD identifies the deepest decline in the sample. Peak-to-trough and trough-to-recovery durations describe its persistence.
- Assumptions and scope
- Asset-price drawdown and strategy-equity drawdown describe different objects. An unrecovered episode has no observed recovery duration.
Related reading: Campbell, Lo & MacKinlay (1997)
VaR validationRolling VaR backtesting · Coverage and independence
Check risk thresholds against subsequent observed losses.
Principle
Each historical VaR uses only preceding returns. Exceedances are examined against expected frequency and independence.
- Interpretation
- Exceedance frequency and coverage tests assess model adequacy over the evaluation sample.
- Assumptions and scope
- Short samples limit test power. Failure to reject a test does not establish future model accuracy.
Related reading: Kupiec (1995) · Rockafellar & Uryasev (2000)
Correlation regimeCorrelation networks · Edge density · Hysteresis
Measure changing connectedness in an asset panel.
Principle
Calculate the share of pairs above an absolute return-correlation threshold, smooth it and apply separate entry and exit thresholds.
- Interpretation
- High density means many pairs are strongly related, including negative relationships.
- Assumptions and scope
- Aligned panel data and defined variances are required. Density does not indicate market direction or a rise probability.
Related reading: Campbell, Lo & MacKinlay (1997)
Asset allocationRisk-based allocation · Hierarchical risk parity
Explore hypothetical allocation across selected assets.
Principle
Estimate risk relationships from aligned returns, then apply the selected optimiser to produce weights.
- Interpretation
- Weights describe a research allocation under its assumptions and estimation window.
- Assumptions and scope
- This is hypothetical long-only allocation, not account holdings or evidence of after-cost performance.
Related reading: Campbell, Lo & MacKinlay (1997)
Hedging researchOrdinary least squares · Historical price sensitivity
Estimate price sensitivity between two assets.
Principle
Regress primary prices on secondary prices with an intercept to estimate sensitivity and residuals.
P_primary,t = a + βP_secondary,t + εₜ
a is the intercept; β is estimated price sensitivity; ε is the residual.
- Interpretation
- Read β in the price units of both series over the observation window.
- Assumptions and scope
- A price coefficient is not a hedge position size; units, basis risk, costs and short-sale constraints must be considered.
Related reading: Campbell, Lo & MacKinlay (1997)
Execution costsSlippage · Linear and square-root impact models
Estimate execution costs under explicit assumptions.
Principle
Apply fixed slippage or size-dependent price impact, then add fees for the modelled trades.
- Interpretation
- Buy price, sell price and round-trip cost are hypothetical quantities around the same reference price.
- Assumptions and scope
- Size-aware models require suitable volume and units. Impact coefficients are not assumed calibrated to the asset.
Related reading: Campbell, Lo & MacKinlay (1997)
Factors, strategies & validation
12 methods
Factor researchCross-sectional factors · Rank information coefficient
Evaluate the relation between factor scores and subsequent returns.
Principle
Compute selected factors on a panel, compare ranks with completed forward returns and examine score groups.
ICₜ = corr(rank(fᵢ,ₜ), rank(rᵢ,ₜ→ₜ₊ₕ))
f is the factor score; r is the observed forward return; i ranges across assets at the same observation.
- Interpretation
- IC and score-group differences diagnose in-sample relationships.
- Assumptions and scope
- Sufficient assets and valid inputs are required. Current diagnostics do not include costs or out-of-sample validation.
Related reading: Fama & French (1993)
Multi-factor researchCross-sectional factor standardisation and ranking
Combine factors into an allocation hypothesis.
Principle
Cross-sectionally standardise momentum, reversal, volatility and volume factors; add equally weighted scores and select the leading group.
- Interpretation
- Read current scores and the held basket alongside the latest ranking time.
- Assumptions and scope
- Suitable panel and volume inputs are required. Ranking rules do not establish alpha or after-cost performance.
Related reading: Fama & French (1993)
Alpha ZooRegistered factor definitions
Explore registered factor definitions.
Principle
Select factor formulas compatible with the market and inputs, then compute scores on the selected panel.
- Interpretation
- Scores express formula values under observed inputs and support research hypotheses.
- Assumptions and scope
- An alpha name is a library label. A formula or score does not establish excess returns.
Related reading: Fama & French (1993)
Seasonality & calendar effectsCalendar-conditioned return averages
Explore return differences across calendar categories.
Principle
Group returns by month and weekday, calculate means and compare with the available seasonal rule.
- Interpretation
- Differences between groups describe historical calendar patterns in the selected period.
- Assumptions and scope
- Small samples, period selection and multiple hypotheses can produce chance findings.
Related reading: Campbell, Lo & MacKinlay (1997)
Pair trading researchCointegration-based pair research
Study relative relationships within an asset pair.
Principle
Examine cointegration, regression with an intercept and spread deviations from their baseline.
- Interpretation
- The spread and standardised deviation help track departures from the estimated relationship.
- Assumptions and scope
- The pair needs aligned data and a stable relationship. Spread research does not authorise shorts in long-only assets.
Related reading: Engle & Granger (1987)
Machine-learning strategyMachine-learning strategies · Walk-forward training
Inspect retained historical evidence for a machine-learning strategy.
Principle
Supported strategies train sequentially through time, score signals and map them to position targets; this method reads matching versioned results.
- Interpretation
- Interpret performance, costs and validation alongside the strategy’s training procedure.
- Assumptions and scope
- A retained strategy result is required. Opening the method does not retrain it, and model scores are not assumed win probabilities.
Related reading: Campbell, Lo & MacKinlay (1997)
Strategy templatesStrategy catalogue and hypothesis selection
Find rule sets suited to the research question.
Principle
Examine catalogue strategies’ definitions, inputs, intervals and execution conditions.
- Interpretation
- The catalogue supports hypothesis selection for subsequent evaluation.
- Assumptions and scope
- Catalogue inclusion does not imply an executed evaluation or verified profitability.
Related reading: Murphy (1999)
Research strategy generationResearch question to executable-rule proposals
Turn a question into strategy proposals for review.
Principle
Generate proposed rules and code for the question, asset and supported execution conditions.
- Interpretation
- Proposals are inputs for code review, validation and subsequent comparison.
- Assumptions and scope
- Generation does not execute code or measure performance; proposals need review and a separate evaluation workflow.
Backtest diagnosisHistorical performance and cost diagnosis
Examine the characteristics and drivers of a backtest result.
Principle
Examine metrics, fees and after-cost performance from a retained run alongside simulation assumptions.
- Interpretation
- Compare returns with drawdown, trading frequency and cost effects when available.
- Assumptions and scope
- Conclusions use the retained run’s data. Opening diagnosis does not rerun it or add out-of-sample validation.
Related reading: Campbell, Lo & MacKinlay (1997)
Strategy validationStored robustness and validation evidence
Assess retained strategy-validation evidence.
Principle
Inspect checks actually present in the result, their data scope and completion status.
- Interpretation
- Distinguish completed checks, insufficient evidence and checks not performed.
- Assumptions and scope
- Walk-forward or holdout evaluation cannot be inferred without corresponding retained evidence.
Related reading: Campbell, Lo & MacKinlay (1997)
Benchmark comparisonAligned strategy and benchmark comparison
Compare strategy performance with a reference alternative.
Principle
Compare retained strategy and benchmark equity over the same observation window.
- Interpretation
- Read return differences alongside risk and the run’s assumptions.
- Assumptions and scope
- Complete aligned benchmark data are required. Costs preceding the first retained observation are not reconstructed.
Related reading: Campbell, Lo & MacKinlay (1997)
Performance attributionPerformance attribution · Factor and allocation effects
Explore components contributing to performance.
Principle
Use original run artifacts to examine factor and allocation attribution when sufficient data are present.
- Interpretation
- Read contributions by component, observation period and the run’s scaling basis.
- Assumptions and scope
- Incomplete attribution inputs produce partial results. Statistical contribution does not establish an economic cause.
Related reading: Fama & French (1993)
Price forecasting
Learning sequence structure to forecast prices
A foundation model specialised for market data learns representations of price sequences during pretraining. An autoregressive Transformer uses temporal relationships to generate a continuation from observed context. The estimated path is retained separately from actual prices, allowing subsequent outcomes and errors to be compared.
Open Price forecasting- Observed price sequence
- Encode representations
- Generate autoregressively
- Decode and compare with outcomes
Foundation modelling for price sequencesAutoregressive Transformer · Learned representations
Estimate a continuation of a price sequence from its observed structure and dynamics.
Principle
Prices are encoded into learned representations. Attention models sequence relationships; autoregressive generation conditions each continuation step on context and preceding generated steps.
- Interpretation
- The forecast is a conditional estimated path. Subsequent observed prices remain separate for error evaluation.
- Assumptions and scope
- Learned patterns may fail under market change. The current price-based path is not a calibrated win probability or confidence band.
Related reading: Vaswani et al. (2017) · Shi et al. (2025)
Scenario simulation
Exploring behaviour and chains of response
Agent-based simulation models participant groups with different objectives, information and constraints. Their responses update shared conditions and influence subsequent rounds. Scenario branches explore transmission paths, sensitive assumptions and observable developments to monitor as conditions change.
Open Scenario simulation- Context and assumptions
- Participants and relationships
- Interaction rounds
- Branches and consequences
Behaviour and interaction simulationAgent-based simulation · Generative agents
Explore chains of responses to changes in information, policy or research conditions.
Principle
Participant groups differ in objectives, information and constraints. Agents use context and prior-round memories to form responses; shared state evolves through interaction rounds.
- Interpretation
- Branches expose transmission paths, sensitivities and triggers. Readers can inspect the reasoning behind each development.
- Assumptions and scope
- Behaviour is modelled under research assumptions. Simulation does not establish empirical causation or a scenario’s occurrence probability.
Related reading: Park et al. (2023)
Interpreting and testing results
Turning ideas into testable rules
For supported assets and engines, strategy ideas become decision rules that can be simulated on historical data. Performance is examined against a benchmark, costs, drawdown and execution assumptions. Results from one period provide a basis for further research into robustness and out-of-sample applicability.
Interpreting measurements in economic context
Financial news, policy, economic calendars and asset information add explanatory context to market movements. Conversational AI helps select tools, synthesise calculated results and explain them in relation to the research question. Economic context is examined alongside price analysis; the current price-sequence forecaster uses price inputs.
Comparing results as markets develop
Statistical analysis describes or tests relationships in data. Forecasting produces a price path for later comparison. Scenarios describe developments conditional on assumptions and modelled behaviour. Keeping these roles clear helps readers reassess a view when new information arrives.
Foundational reading
16
The following research and texts provide theoretical foundations for the methods described.
- Murphy (1999)Technical Analysis of the Financial Markets (opens in a new tab)
- Campbell, Lo & MacKinlay (1997)The Econometrics of Financial Markets (opens in a new tab)
- Dickey & Fuller (1979)Distribution of the Estimators for Autoregressive Time Series with a Unit Root (opens in a new tab)
- Bollerslev (1986)Generalized Autoregressive Conditional Heteroskedasticity (opens in a new tab)
- Engle & Granger (1987)Co-integration and Error Correction: Representation, Estimation, and Testing (opens in a new tab)
- Rockafellar & Uryasev (2000)Optimization of Conditional Value-at-Risk (opens in a new tab)
- Künsch (1989)The Jackknife and the Bootstrap for General Stationary Observations (opens in a new tab)
- Vaswani et al. (2017)Attention Is All You Need (opens in a new tab)
- Shi et al. (2025)A Foundation Model for the Language of Financial Markets (opens in a new tab)
- Park et al. (2023)Generative Agents: Interactive Simulacra of Human Behavior (opens in a new tab)
- MacKinlay (1997)Event Studies in Economics and Finance (opens in a new tab)
- Kahneman (2002)Maps of Bounded Rationality: A Perspective on Intuitive Judgment and Choice (opens in a new tab)
- Fama & French (1993)Common Risk Factors in the Returns on Stocks and Bonds (opens in a new tab)
- Embrechts, Klüppelberg & Mikosch (1997)Modelling Extremal Events: for Insurance and Finance (opens in a new tab)
- Nelsen (2006)An Introduction to Copulas (opens in a new tab)
- Kupiec (1995)Techniques for Verifying the Accuracy of Risk Measurement Models (opens in a new tab)
