Every trader who has lost money on options has, at some point, seen a beautiful backtest. A curve that climbs from left to right. A win rate that looks like a school report card. And then real money goes in, and the curve does not repeat.
That gap has a boring, unglamorous cause, and it is almost never the strategy. It is the fill — the actual price at which your order met a real buyer or seller. Backtests assume fills. Markets negotiate them.
In the nets, the bowler has no slip cordon. Nobody sets a field for you. There is no scoreboard and no crowd. You can middle twenty balls in a row and it means something — but it does not mean you scored a fifty.
A backtest is net practice. It replays prices that already happened and asks a simple question: if I had wanted to buy here and sell there, what would I have made? The word doing all the damage in that sentence is wanted. Wanting to sell at ₹268 is not the same as somebody being willing to buy from you at ₹268.
1. The spread. A SENSEX weekly option does not have one price. It has a bid (what buyers offer) and an ask (what sellers demand). If the bid is ₹244.60 and the ask is ₹246.40, that ₹1.80 is not a fee anyone charges you — it is simply the cost of being in a hurry. On 20 quantity per lot, a ₹1.80 spread is ₹36 per lot, each way. Most backtests quietly fill at the midpoint, which is a price neither side actually offered.
2. Slippage on the way out. Entries are usually the honest part — you choose when to enter, and you can wait. Exits are where it hurts, because a stop-loss fires precisely when everyone else also wants out. That is the moment the book is thinnest.
3. The gap between ticks. This is the one almost nobody models, and in our own record it was the largest of the three.
We run a SENSEX level-break method: map support and resistance, take only confirmed breaks, fixed stop-loss, a scale-out, and a trailing floor. For a stretch in August we were measuring the trailing floor against our own model and congratulating ourselves.
Then we compared modelled exits against the prices the exchange actually printed. In one week, six out of six trailing exits filled below the floor we thought we had locked. Not because the logic was wrong — because our price updates were arriving up to eight seconds apart, and in a fast tape eight seconds is an eternity. On average, real fills came in about 8.2 points worse than the model on fast moves.
The method did not change. The honesty of the measurement did.
| Date | Session P&L |
|---|---|
| 1 Sep | −₹19,350 |
| 3 Sep | +₹38,723 |
| 4 Sep | +₹3,600 |
| 8 Sep | +₹11,192 |
| 9 Sep | +₹25,287 |
| 10 Sep | −₹26,200 |
| 11 Sep | +₹5,723 |
| Month to date | +₹38,975 |
Two red sessions out of seven, one of them nearly the size of the best green one. That is what the record looks like when losses are printed at the same font size as wins. Past results describe what happened; they do not describe what will happen.
The uncomfortable summary is this: most strategies that "stop working in live" never worked. They worked in a version of the market where every price you wanted was available to you, in size, instantly, for free. That market does not exist. The sooner a backtest is made to trade in the real one, the sooner it stops flattering you.