Kadvari's signal uses each day's closing price to trigger a trade executed at that same close. Re-run with execution at the next open and the annual return falls from 15% to 4%. Where did the 11 points go, and how do you fix the timing?

The 11 points are look-ahead: the backtest buys at the very close that triggered the signal, a price no order could have reached, and pockets the move from that close to the next open. Over 50 trades the overnight gap at entry is worth about 12.5 points, partly given back at exit. Fix the timing by trading at the next open, or by computing the signal from a price before the close and testing a fill in the closing auction, then add costs.

Imagine betting on a cricket match with a bookmaker who accepts the bet after the last ball. Anyone could win that way, and no bookmaker allows it. A closing price is known only when trading has finished, so a signal that needs the close cannot also be executed at the close. This is a look-ahead error, and it is easy to make because in a daily data file the close and the signal sit on the same row, one column apart. The backtest is, in effect, placing the order after seeing the result.

Follow one typical trade. On Day 1 the stock closes at 100.00, above its 20-day high, and the signal fires. The backtest buys at 100.00. A live system learns of the signal after the close and buys at Day 2's open, 100.25. On Day 2 the close of 100.30 triggers the exit; the backtest sells there, the live system at Day 3's open, 100.33. The backtest earns 0.30%, the live trade 0.08%: the 0.25 move overnight after the signal is exactly the part of the trade no one could have caught. It is not random either: a strong close that breaks a 20-day high tends to be followed by a higher open, because other traders read the same close overnight.

Scale it up. With 50 trades a year, entry at the next open costs 50 x 0.25 = 12.5 points, and exiting at the next open instead of the close happens to give back 1.5, because the drift continues a little after the exit signal too. Net, 15% becomes 4%: most of the backtest's return was the overnight move the signal could not have traded.

There are two honest designs. Either compute the signal at the close and trade at the next open, which is what the reviewer did, or compute it from a price available before the close, say 20 minutes earlier, and test a fill in the closing auction at the real closing price. The second keeps more of the overnight move but changes the signal, since the early price is not the close, so it must be backtested as its own strategy. Either way, add trading costs, which for a 4% strategy with 50 round trips a year are a large share of what remains. Then sweep the rest of the code for the same mistake: adjusted prices that use later corporate actions, index membership as of today, and fundamentals stamped with their period end rather than their release date.