The Overnight Drift

Every trading day has two separate legs, and they behave nothing alike.

Take the simplest possible test. On any given day the index opens at 09:15 and closes around 15:29–15:30. Between those two prints, run one trade: buy the open, sell the close. Call that the session leg — pure intraday exposure, no overnight risk carried.

Now do the mirror image. Buy at today's close, sell at tomorrow's open. Call that the gap leg — no intraday exposure at all, just whatever happened between the last print and the next one, while the exchange was shut.

Chain each leg through every session for over a decade — 2,856 trading days on Nifty and Bank Nifty, from January 2015 to August 2026 — compounding only within that leg, and see which one actually built the return.

Nifty: the session was a net loser. The gap was the entire story.

The two legs recombine exactly — ₹14.20 × ₹20.64 (indexed) reproduces the actual buy-and-hold line to the last decimal, because that is arithmetically what a daily close and a daily open decompose into. This isn't an approximation of the real return; it's the real return, split into its two component parts.

Read the session column carefully — it is not merely "smaller than the gap," it is negative. A pure buy-the-open-sell-the-close strategy, held for eleven and a half years with no other edge, lost money on both indices. Every rupee of the long-run gain — and the rupees needed to cover the session's loss besides — came from the close-to-open leg.

The same story in raw index points, not compounded percentages.

Percentages compound, which is the right way to measure a return but can obscure the arithmetic. So set compounding aside and just add up the points: every day's close − open summed across the whole period is the total the session leg contributed; every day's open − previous close summed the same way is what the gap contributed. The two sums add up exactly to the index's total point gain over the period — no rounding, no approximation.

This is the plainest way to say it: if you'd captured only the overnight points on Nifty, you'd have banked 2.5 times the index's actual eleven-year gain — the session then handed roughly three-fifths of that back. On Bank Nifty the overnight haul is 1.8 times the net gain. Either way, the number the market reports as "the index went up 16,000 points" is really "the index went up 40,000 points overnight and gave back 24,000 of them during the day."

The gap was positive in every full calendar year measured. The session mostly wasn't.

A decade-long average can hide a story that's really about two or three outlier years. Splitting by calendar year rules that out here.

The session leg is messier, and worth stating honestly rather than rounding off. On Nifty it was negative in ten of the eleven complete years and essentially flat in the eleventh. On Bank Nifty it was negative in nine of eleven, but genuinely positive in 2022 and 2025 — so "the session always loses" is an overstatement for Bank Nifty specifically, even though it's a fair description of the eleven-year average. 2026 is the year to watch: it's the first year in the entire sample where Nifty's gap leg has gone negative alongside its session leg, and the first year Bank Nifty's session has run positive while its gap ran negative — worth returning to once the year is complete rather than reading much into eight months.

This isn't an India quirk. It's one of the oldest documented anomalies in equity markets.

The same pattern shows up wherever anyone has looked. Academic work on the S&P 500 going back to the early 1990s — Cooper, Cliff & Gulen (2008) and, on ETF data a few years later, Kelly & Clark (2011) — found that essentially the entire long-run US equity risk premium was earned overnight, with the intraday session contributing close to nothing on average. More recent tallies of SPY over the last three decades land on numbers the same shape as the ones above: roughly flat intraday, and almost all of the index's growth captured close-to-open.

A few explanations get offered, and they're complementary rather than competing:

• Information doesn't wait for the opening bell. In India specifically, GIFT Nifty (formerly SGX Nifty) trades close to 21 hours a day and continuously reprices against US and Asian markets while NSE is shut. By 09:15, a large share of the overnight news — the US close, crude, the dollar, an Asian session already in progress — is already reflected in where the futures are sitting. It shows up as a gap, not as something that unfolds during the day.
• Someone is compensated for holding risk they can't exit. A position held overnight can't be closed if bad news lands at 2am. Market-makers and other intermediaries who warehouse that inventory risk are, in standard market-microstructure explanations, compensated for it — and that compensation shows up as a structural premium on the close-to-open return, distinct from a premium for holding risk during hours when the position could be exited at any moment.
• A lot of intraday flow works against the opening print, not with it. VWAP execution, opening-range fade strategies and end-of-day rebalancing all tend to unwind part of the overnight gap during the session rather than extend it. That's a documented pattern more than a settled causal story, but it's at least directionally consistent with a session leg that drifts negative on average rather than just flat.

A clean decomposition, not a strategy.

It's tempting to read "buy the close, sell the open, repeat 2,856 times" as a trading idea. It largely isn't one, for reasons worth being upfront about:

• You can't actually do this on the cash index. The Nifty and Bank Nifty numbers above are index levels — there's no instrument that lets you buy the index itself at 15:29 and sell it at 09:15 the next day. The nearest real trade is holding index futures overnight, which brings its own funding/carry cost and a roll every expiry — not free, and not modeled here.
• Costs are not in these numbers, and they matter a lot at this frequency. Capturing the gap leg means roughly 2,850 separate overnight positions across the period — brokerage, STT and slippage on that many round trips erode a real edge meaningfully. Independent work testing this exact anomaly on other markets — see Alpha Architect's review — found that realistic trading costs eat most or all of the apparent excess return once you actually try to trade it.
• The premium is compensation for gap risk, not a free lunch. Earnings, RBI and Fed announcements, and geopolitical news cluster overnight precisely because that's when markets are closed and can't reprice continuously. The extra overnight return is generally understood as pay for bearing that risk, not evidence the market is inefficient.

What the numbers do show: on both indices, over more than a decade, the session and the gap behaved very differently. For someone holding the index, the hours the market was open added little or nothing, and on Nifty that part was negative in ten of the last eleven complete years. A long position that must be closed by the end of the day gave up the part of the return that did the work. The study did not test short trades, or intraday signals that pick their days, so it says nothing yet about those.