How to Set Realistic Backtest Assumptions

Why assumptions decide the result

A backtest is a model, and every model is built on assumptions. Most backtesting disappointments do not come from a flawed strategy idea; they come from optimistic assumptions that quietly inflate the result. Assume zero costs, perfect fills at ideal prices, and instant execution, and almost any strategy can be made to look profitable. The market charges for all of those things. The discipline is to make your simulation pay the same prices reality will, so the result you see is a result you could actually have achieved.

Trading costs and slippage

Start with the costs that hit every trade. Commissions are the easy part, since they are usually known and fixed. The harder, larger costs are the bid-ask spread and slippage. The spread is the gap between the price to buy and the price to sell; you cross it on entry and exit. Slippage is the difference between the price you expected and the price you actually got, and it grows with order size and shrinks with liquidity. A frictionless backtest ignores all of this and is therefore describing an impossible market.

A worked example shows the stakes. Suppose a strategy trades 150 times a year and shows a 14 percent gross return assuming no costs. If realistic round-trip friction is 0.12 percent per trade, that is roughly 18 percent of return given away to costs over the year, turning an apparent winner into a loser. The fix is not clever; it is honest. Charge plausible costs, and if anything, charge slightly more than you expect, because reality rarely undershoots.

Fills, timing, and data quality

Next, be honest about how and when trades fill. Assuming you always buy at the day's low and sell at the high is fantasy. A safer default is to fill at the next available price after a signal, not the same instant the signal fired, since you cannot trade on information you did not yet have. This also guards against look-ahead bias, the subtle error of letting your strategy use data that would not have been available in real time. Even a small look-ahead leak can manufacture a spectacular, completely fake edge.

Data quality is the foundation under all of this. Survivorship bias, where your historical universe quietly excludes companies that failed or were delisted, makes the past look safer and more profitable than it was. Use data that includes the losers. Account for dividends and corporate actions where relevant. If your data is wrong, no amount of careful cost modeling will save the result.

How to choose conservative defaults

When you are unsure of the right number, lean pessimistic. Round costs up, fills toward the worse price, and execution toward slower. The reason is asymmetric: if your conservative backtest still shows an edge, you have a margin of safety, and live trading may pleasantly surprise you. If you used optimistic assumptions and the edge was thin, live trading will unpleasantly surprise you instead, with real money. A strategy that only survives under generous assumptions is telling you it will not survive at all.

Common mistakes

The most common mistake is assuming zero or trivial costs. The second is filling at idealized prices that no real order would receive. The third is look-ahead bias that lets the strategy peek at the future. The fourth is using clean, survivorship-biased data that hides the failures. The fifth is tuning your assumptions until the result looks good, which is just another form of overfitting. Realistic assumptions will make your backtest look worse, and that is the entire point: a backtest that flatters you is a backtest that will betray you.

Honest assumptions are not an afterthought in TRION, they are the workflow: you set explicit costs, slippage, and fill rules, read the compiled logic line by line, and run the whole thing in paper mode so a flattering result can never quietly mislead you.

Simulation-only. No broker, no real orders, no promise of profit. AI assists, TRION validates, humans decide.