Every term is defined twice: once the way a filing would put it, and once the way somebody would explain it to you across a table. The second one is usually the one that sticks.
Showing 11 terms
Applying a trading rule to historical data to estimate how it would have performed.
An experiment where you already know the answer, run by someone who wants a particular result. Useful if you are careful and dangerous if you are not.
Using information in a backtest that was not actually available at that point in time.
Trading on a quarterly result weeks before it was filed. Makes any strategy look brilliant.
Using information in a backtest that was not available at the time the decision would have been made.
Buying at today’s close because of a signal that only existed once today finished. It roughly doubles backtested returns and is invisible unless you check.
Estimating the difference between the price a backtest assumes and the price a live order actually fills at.
A daily-timeframe system loses relatively little to it. An intraday one can lose its entire theoretical edge.
Whether a price series is accurate and consistently adjusted.
A common and silent reason backtests look remarkable and cannot be reproduced live.
Getting an intended order actually filled, and the gap between the price a strategy assumed and the price it achieved.
Where backtests go to die. Fills at the next open rather than the close, timeouts, rate limits and costs all run in the same direction, and the damage grows with frequency.
The order in which prices were reached inside a bar — information the bar’s four numbers do not carry.
A candle records how far the session reached each way and throws away when. Any rule with a stop and a target depends on the order, and a backtest has to assume one.
Testing a rule once on data that was not used to develop it.
The only backtest result worth much. Develop on two-thirds, test once on the rest, and resist the urge to tweak afterwards.
When an indicator or chart marker changes its historical values as later data arrives, so what it shows today is not what it showed at the time.
Anything that centres a calculation, or that marks a swing point only once the swing is confirmed, cannot have known the answer when it appears to. A backtest that reads those values as though they were available live produces results you can never reproduce with money.
Averaging price data to suppress short-term noise, as Heikin-Ashi does by blending each bar with the one before it.
It buys clarity by discarding information, and it flatters backtests badly — because the whipsaws it removed are exactly the ones that would have stopped you out live.
Developing rules on one period of data, freezing them, then testing once on data never examined.
The strongest single defence against fooling yourself with a backtest.