Quant Memo
Core

Tracking Error Against the Backtest

Measuring how far a strategy's live returns drift from the returns its own backtest would have produced over the same period, as a running check on whether the live implementation matches the design.

Prerequisites: Sharpe Ratio

Once a strategy goes live, the backtest doesn't stop existing — it can keep running in parallel, fed the same live market data, producing a hypothetical return series to compare against what actually happened in the account. Tracking error against the backtest is the standard deviation of the difference between those two return series: live minus backtest, day by day. A small, stable tracking error means the live implementation is doing what the research said it would do. A tracking error that is large, or growing, means something about the live system — execution, data, position sizing, timing — has drifted away from the design, even if the strategy is still profitable.

This is a different question from "is the strategy making money." A strategy can be having a great month in live trading while its tracking error against the backtest is quietly blowing up, because the live P&L is coming from something other than the intended signal — a slippage effect, a data feed discrepancy, or a risk limit forcing different position sizes than the backtest assumed. Conversely a strategy can be having a mediocre month with tight tracking error, meaning it is behaving exactly as designed and simply hit a rough patch the backtest itself would have predicted. Separating "is it working" from "is it behaving as built" is the whole point of tracking this number.

A concrete example: a pairs-trading strategy backtests with daily returns closely following its historical pattern, and for the first three months live, tracking error runs around 15 basis points a day — consistent with reasonable execution noise. In month four it jumps to 60 basis points a day. Investigating position-by-position, the team finds a data vendor switched a ticker's adjusted close methodology, causing the live signal to compute slightly different entry thresholds than the backtest's historical prices would have implied. The elevated tracking error was the first symptom; the root cause was a silent data change, not a failure of the trading idea itself.

Tracking error against the backtest measures how far live returns diverge from what the same strategy, fed the same live data, would have produced on paper — separating "is the strategy behaving as designed" from the separate and sometimes misleading question of "is it making money right now."

Related concepts

Practice in interviews

Further reading

  • Grinold and Kahn, Active Portfolio Management, ch. 17
ShareTwitterLinkedIn