Survivorship bias: the backtest told by the survivors

A friend forwarded me a backtest last week. Beautiful curve. A basket of fifty large caps, run over ten years, and the line never had a lost decade. His question was simple. Why not just buy this.

So I asked which fifty names. They were today's fifty. And the market he was testing them against was the last ten years. That is the whole problem, and it has a name.

An index is not a fixed list. It gets reviewed twice a year, and names get swapped out. Roughly twenty five large caps cycled through the list over the last decade. Here is the part that matters: a name usually leaves the index after it has already shrunk or de-rated. It leaves after it has already hurt the people who held it.

Now run your backtest on today's fifty names. You are quietly holding only the companies that survived all the way to today. The ones that got deleted, often after halving, never show up. Not because the test is dishonest, but because they are not on the list you copied off the screen.

Same ten years, two different universes. The survivor list flatters the curve by something like a couple of percentage points a year. That gap is not skill. It is the losses you edited out.

The deletions are not noise. They are the data. They are the exact drawdowns you would have lived through if you had actually been in the market the whole time. Cut them and you are not measuring a strategy. You are measuring hindsight.

So before you trust any backtest, ask one question. Did it use today's members, or the index as it actually stood on each date. And does the strategy say what it buys and sells when a name falls out, or does it just assume that name was never there.

Every backtest is a story told by the survivors. The honest one keeps a seat for the names that did not make it.

How a name leaves the list

Index membership follows a calendar. NSE reviews its broad indices twice a year, on market data for the six months to the end of January and the end of July, and the changes take effect at the end of March and the end of September. A company that has shrunk relative to the others drops in the ranking and makes way for one that has grown.

1. Month 1Company A is a member

   It sits in the index at its full weight and in every fund that tracks it.

2. Months 2 to 6The business stumbles

   The stock halves. Every index investor holds it all the way down.

3. ReviewThe ranking is checked

   Company A's market value now ranks below the cut off on the review data.

4. Effective dateCompany A is replaced

   It leaves after the fall. A company that has risen takes its seat.

Run a backtest years later on today's list and Company A is simply absent. Its fall, which every real investor in the index lived through, has been deleted from the history. The company that replaced it is present, and present from the start of the test, long before it was large enough to be chosen.

Lists change more than people think

The middle number is a small, useful reminder. An index called the Microcap 250 held 254 names at the end of September 2026, because changes between reviews and corporate events move the count. A list that looks fixed in its name is moving underneath.

What two points a year does to a backtest

The opening puts the survivor flattering at about two percentage points a year. That sounds small. Compound it and it is not.

On Rs 10 lakh over ten years, the survivor version shows about Rs 5.1 lakh that no real investor ever earned. Over twenty years the gap grows to about Rs 29 lakh. And because the deleted names usually fell hard before leaving, the survivor test also shows a shallower worst fall, which is the number most people use to decide whether they can hold a strategy.

A test you can trust

1. Use the list as it stoodOn every rebalance date, pick only from the index members on that date.
2. Keep the leaversHold a deleted name until the rule sells it, at the price it actually traded.
3. Include the deadMerged, suspended and delisted companies stay in the data with their final prices.
4. Charge real costsEvery trade pays costs and tax, as it would in a real account.

The free Learn module on how backtesting works covers survivorship, look ahead bias and overfitting with examples and a quiz. The CAGR calculator turns any two values into an annual rate, which helps when you check a backtest's claims by hand. For another way a clean looking result can mislead, read the optimizer is an error maximizer.