Five ways a backtest lies to you

In rough order of how often I've caught myself: 1. Look-ahead. Using a value the bar hadn't finished producing yet. The most common and the most flattering. 2. Survivorship. Testing on the names that still exist today.

3. Optimistic fills. Assuming you got the price you saw.

4. One-outlier profits. Remove the best trade; if the edge disappears, you don't have an edge, you have an anecdote. 5. Too many knobs. Every parameter you tune is another chance to fit the noise. None of these are exotic. They're just easy to not look for when the equity curve is pretty.
BacktestBetty
16 posts · 0 followers
+ Follow

Related reading

4 replies

DataDrivenDee· Jul 2026 ago
Number 4 should be a default check, not something you do when suspicious. I run it on everything now.
G
GapHunterMike· Jul 2026 ago
number 1 killed a strategy i traded for four months. felt great right up until it didn't.
C
CoffeeAndCharts· Jul 2026 ago· edited Aug 2026 ago
saved this one to read properly with my coffee, glad i did. my own version of #5 is that i only let myself change one knob per morning, otherwise by friday i have no idea which change did what. also #3 got me on thin premarket names, the price i saw and the price i got were basically strangers.
Q
QuietVol· Aug 2026 ago
One I would add to the list is sample size, which quietly sits underneath number 4. Thirty trades feels like data and it is really just a wide distribution wearing a suit — the spread of possible outcomes around that mean is enormous. Related to your number 5: each parameter effectively costs you degrees of freedom, so the more you tune the more trades you need before the result means anything. Do you have a rough minimum trade count you require before you take a curve seriously?
Sign in to reply →
← All strategies & backtesting