Key takeaways
- Historical performance is a record of the past, not a forecast of the future.
- Edges decay because participants change, microstructure evolves, and costs change.
- Indian markets have seen multiple structural breaks in the last decade.
- Backtests are educational evidence, not predictions.
Why backtests are not forecasts
Historical performance — even an honestly measured one — is a record of what happened, not a forecast of what will. The same rule applied to the future produces a different outcome because the inputs themselves change: participants, microstructure, costs, regulation, and the macro environment.
Regime shifts
Markets move through regimes — low and high volatility, trending and mean-reverting, retail-driven and institution-driven. A rule that produced a clean curve in one regime can break down in another. A backtest crossing only one regime tells you about that regime, not about the strategy in general.
Microstructure changes
Indian markets have seen meaningful structural changes in the last decade: weekly expiries appeared on multiple indices, lot sizes were revised, peak-margin rules tightened, STT rates were restructured, and retail participation shifted significantly. A backtest that spans these changes needs to be read with care; many historical 'edges' were artefacts of the prior microstructure.
Crowding and decay
Edges decay when participants discover them. The same rule, once visible and traded by many, generates less or no edge in the future. Published research is, by definition, visible.
Data quality
Historical data has its own pathologies: missing prints, revised figures, vendor differences, and reconstructed series. A backtest is only as good as the data underneath it.
Common mistakes
- Treating historical CAGR as expected forward CAGR.
- Ignoring structural regime breaks inside the data window.
- Assuming participant composition is constant.
- Treating backtest evidence as a forecast rather than as research.
How this appears in OptionScience reports
Reports include a Limitations section so the author can name the structural reasons the study may not extrapolate.
Practical educational example
Checklist
- Does the data window cross a known structural break?
- Are limitations written down by the author?
- Is the study presented as research, not as a forecast?
- Is participant composition assumed to be constant?