A benchmark is the alternative you would have held if you had done nothing clever. Every relative number on the analytics screen depends on it: Excess Return, Alpha, Beta, Tracking Error and Information Ratio are all computed against whichever benchmarks you selected in the backtest config, over the dates the two series share.
Choose it before you run the backtest, not after you see the result. Picking the benchmark your strategy happens to beat is the oldest way to fool yourself with real data.
Matching the benchmark to the universe
| Your universe | A defensible benchmark |
|---|---|
| Nifty 50 constituents | Nifty 50 TRI |
| Broad large and mid-cap | Nifty 100 TRI or Nifty 200 TRI |
| Nifty 500 stock selection | Nifty 500 TRI |
| Small-cap screens | Nifty Smallcap 250 TRI |
| Single-sector rotation | The matching sector index, plus the broad index as a second line |
| Sector or thematic ETFs | An equal-weight basket of the same ETFs, held passively |
Benchmarking a small-cap strategy against the Nifty 50 is the most common self-deception in Indian retail backtesting. In a small-cap bull run it produces enormous apparent alpha that is entirely a size premium.
Price index versus total return index
A price index excludes dividends. A total return index reinvests them. Indian large-cap dividend yields have generally run around 1% to 1.5% a year, so benchmarking against a price index quietly credits you with roughly a point of annual outperformance you did not earn. Use the TRI variant wherever it is available, and note which one you used.
Caveats
Relative metrics are computed only over overlapping dates. A benchmark series that starts later than your backtest, or has gaps, shortens the comparison window without warning. Check the date range before reading the alpha.
The benchmark itself is not a passive object. Index constituents change, and a strategy compared against an index that quietly rotated into the same winners is not being tested against a genuine alternative.
One benchmark is rarely enough. A strategy worth deploying should beat both a broad index and the obvious cheap substitute, such as an equal-weight version of its own universe.