The honest backtest: costs, slippage, settlement & Deflated Sharpe for NSE/BSE index options

Why most retail options backtests overstate returns

Four silent inflators, in rough order of damage:

1. No or frozen costs. A NIFTY iron condor turning over four legs twice a day pays real STT, brokerage, exchange charges, GST and stamp duty. Omit them and a break-even strategy looks profitable.
2. Fantasy fills. Backtests that fill at the mid (or even the bar close with zero slippage) ignore the bid-ask spread you actually cross — wide on options, wider at the open and the close.
3. Wrong expiry handling. NSE/BSE index options are European, cash-settled. An in-the-money leg held to expiry settles against the closing index value, not the last traded premium — and out-of-the-money legs expire at exactly zero.
4. A headline Sharpe that lies. Try enough variations and one will look great by luck. A raw Sharpe ratio doesn't account for how many you tried.

The fix for each follows.

The real Indian index-options cost stack (2026)

Costs are charged on the option premium turnover (not the contract notional), in this order of operations:

Key correctness points an honest engine must get right:

• Rates are point-in-time — a 3-year backtest spans multiple STT/txn regimes; freezing today's numbers is wrong.
• STT is sell-side on premium, and applies to intrinsic value on exercised ITM options.
• GST's base excludes STT and stamp.

A realistic cost model typically slices 0.5–3% per year off naive returns, and far more for high-turnover, 4-leg strategies.

Realistic execution: slippage, liquidity, settlement

• Slippage / spread. Don't assume mid-price. Model slippage as a fraction of the bid-ask spread that scales with the number of legs (a single naked leg crosses more of its spread than a packaged 4-leg condor), and widen it in high-volatility regimes. Cap order size relative to traded volume (~5% of ADV) so large orders don't fill at a single price.
• The long-only trap. The most popular Indian retail option strategies are credit / short (short straddles, iron condors, credit spreads). A backtester that can only model buying options silently skips the sell legs — leaving a credit strategy with no position and a meaningless result. An honest engine either models the short side (with a margin reserve) or clearly flags the result as unrepresentative.
• Expiry settlement. At expiry, settle ITM legs at intrinsic value vs the closing index, expire OTM legs at zero, and never carry a position across the expiry roll (which otherwise books phantom profit by marking an old entry against a new contract).

The Deflated Sharpe Ratio (DSR): why raw Sharpe overstates edge

The Sharpe ratio measures return per unit of risk — but a raw Sharpe says nothing about how many strategy variations you tried to find it. Bailey & López de Prado showed that the expected maximum Sharpe across N independent trials is strictly positive even when the true edge is zero: try ~10 variations and you'd expect a backtested Sharpe near 1.5 from pure luck.

The Deflated Sharpe Ratio corrects for two things a raw Sharpe ignores:

1. Multiple testing — it subtracts the Sharpe you'd expect from the best of N trials by chance (the more you searched, the higher the bar).
2. Non-normal returns — it penalises negative skew and fat tails (common in short-option strategies, which win small often and lose big rarely).

Practical takeaway: track how many variations were tested and judge the deflated number, not the headline. A 2.5 Sharpe found among 100 tries can deflate below the 1.0 “real edge” bar.

Probability of Backtest Overfitting (PBO)

PBO estimates the probability that the configuration which looked best in-sample will underperform the median out-of-sample — i.e. that your “winner” is curve-fit. It's computed via Combinatorially Symmetric Cross-Validation (CSCV): repeatedly split the history, pick the in-sample best, and check how often it disappoints out-of-sample. A customary bar is PBO < 5%; above that, treat the result as likely overfit rather than a real edge.

Lot sizes changed (Jan 2026)

NSE rebaselined index F&O lot sizes effective Jan 2026: NIFTY 75 → 65, BANKNIFTY 35 → 30, SENSEX stays 20. A backtest using stale lot sizes mis-states position size and per-lot P&L.

How to read a backtest verdict honestly

Before trusting a number, ask:

• Were realistic costs (the full stack above) applied, point-in-time?
• Were fills modeled with slippage, not mid-price?
• For a credit strategy, were the short legs actually modeled?
• Is the headline a deflated Sharpe, and what was N (variations tried)?
• Is PBO below ~5%?
• Was the strategy tested out-of-sample / across regimes, not just one lucky window?

If a tool can't answer these, its returns are decoration.

Where this is built in

Algoshastra's Shastra engine applies the full Indian cost stack point-in-time, models slippage and European cash-settlement, supports credit/short strategies, and reports a Deflated Sharpe and Probability of Backtest Overfitting on every verdict — so the number you see is the number you'd actually have lived. You describe a strategy in plain English, it backtests honestly, and you export the verified strategy to run on your own broker.

Common questions

Read next

The honest frame

Investment in securities market are subject to market risks. Read all the related documents carefully before investing.

Backtested results are hypothetical, do not represent actual trading, and are not indicative of future results. This article is educational and is not investment advice or a recommendation; Algoshastra is a strategy-building and testing tool, not a registered investment adviser or research analyst. Past or backtested performance does not guarantee future returns.