NIFTY algo strategies, and what each one is really being paid for.

01What makes a strategy an algo

The word "algo" gets used loosely. For our purposes it means something narrow: every decision is determined by written rules, and no human judgement enters at trade time. Given the same market data, the system produces the same action every time.

That property is not a stylistic preference. It is what makes the strategy testable. If a rule says "enter when momentum looks strong," you cannot backtest it, because "looks strong" resolves differently on different days and in hindsight it always resolves in your favour. If the rule says "enter when the index trades 0.4% above the session VWAP between 09:45 and 11:00," you can run that against four years of tick data and get an answer you did not choose.

A complete NIFTY algo specifies at minimum: the entry condition, the instrument and strike selection rule, the position size, the stop-loss level and where it lives (in your code or on the broker's book), the exit condition, and the behaviour when something goes wrong: a missed fill, a disconnect, a stuck order. That last category is where most home-built systems are thin, and it is the part that costs real money.

02The four strategy families

Strip away the branding and almost every systematic NIFTY options strategy belongs to one of four families. They are distinguished by what the strategy is being compensated for.

Notice that the "fails when" column is not optional. A strategy family without a known failure mode has simply not been tested against one yet.

03Short volatility, the dominant family

Most Indian index option algos are, underneath, short volatility. The structural claim is that implied volatility on NIFTY options tends to price slightly above the volatility that subsequently gets realised, because there is persistent demand for downside protection and a limited supply of people willing to sell it.

That premium is real, and it is also the most crowded trade in the Indian market. Being right about the premium is not the hard part. The hard part is surviving the days when realised volatility overshoots implied by a wide margin, because the loss distribution is deeply asymmetric: you collect small amounts often and pay a large amount rarely.

The practical response is hedging. Buying a further out-of-the-money option against the one you sold converts an open-ended loss into a defined one. It costs part of the premium, which lowers the headline return, and it is the single biggest difference between a strategy that survives a volatility event and one that does not.

04Intraday mean reversion

The second family trades the tendency of the index to revert toward a reference price within a session. The reference is usually a volume-weighted average price, an opening range, or a rolling band. When price extends unusually far from that reference without a corresponding change in volume or breadth, the strategy takes the other side.

What you are being paid for here is liquidity provision. Someone needed to transact urgently, they pushed price away from fair value, and you absorbed it. The edge per trade is small and the hit rate is high, which makes this family feel comfortable and makes its failures surprising: when the move away from the reference is driven by genuine new information, reversion does not come, and a high-hit-rate system takes an outsized loss against a run of small wins.

Reversion strategies are also the most sensitive to execution quality. If your edge is 0.15% per trade and your slippage is 0.2% per fill, you do not have a strategy, you have a fee-generation machine. This is why we insist on modelling slippage on both sides of every trade when backtesting.

05Expiry-day theta capture

NIFTY now has weekly expiries, which means there is an expiry-day or near-expiry session available most weeks. On those sessions, the time value remaining in an option decays to zero over a handful of hours. A strategy that is net short options through that window collects the decay.

The appeal is that the position opens and closes within a single session, so nothing is carried overnight and the gap between one day's close and the next day's open, historically the source of the largest single-day losses in Indian option selling, is removed from the risk picture entirely.

The risk does not disappear, though. It relocates. On expiry day the option's sensitivity to price movement rises sharply as expiry approaches, so a move that would be unremarkable on a Monday can be severe in the final hour of an expiry session. Removing overnight risk buys you a great deal; it does not buy you safety. We cover the mechanics in 0 DTE and 1 DTE options.

06Trend and breakout

The fourth family is the mirror image of mean reversion: it assumes that a decisive move away from a reference price continues rather than reverts. On NIFTY, pure trend-following on an intraday horizon has historically been difficult, because the index spends a large share of sessions in a range and a breakout system pays a small cost on each false signal.

Where the family earns its place is as a hedge on the rest of the portfolio. Trend systems tend to perform best exactly when short-volatility systems perform worst, during sustained directional moves and volatility expansion. Running a small trend sleeve alongside a larger premium-selling book is a structural diversification rather than a return enhancement, and it should be judged on that basis.

07How each family fails

The failure modes are worth stating plainly, because they are what position sizing has to be built around.

• Short volatility fails on a discontinuous move. Not a large move: a fast one, where the price gaps through your stop level and you are filled well beyond where you intended. This is why a stop-loss resting on the broker's book behaves differently from a stop-loss evaluated in your own code, and why spike detection matters.
• Mean reversion fails on information. The system cannot distinguish a dislocation caused by a large order from one caused by news, and it adds to the losing side of the second case by design.
• Theta capture fails on expiry-day direction. Sensitivity is highest exactly when there is least time left to recover.
• Trend fails on chop. It is a slow bleed rather than a shock, which makes it psychologically harder to hold through, and which is why discretionary intervention destroys trend systems more often than market conditions do.

08Telling a real edge from a fit

Any parameter search over enough configurations will produce something that looks spectacular in hindsight. The question is whether the result survives contact with data it was not chosen on.

In one SENSEX-focused search we ran across roughly 3.2 million configurations, about 20,000 cleared a walk-forward gate. We then checked whether ranking well in-sample predicted ranking well out-of-sample. The Spearman rank correlation between the two was −0.001, statistically indistinguishable from zero.

That number is the whole lesson. The in-sample score had no predictive power for out-of-sample performance. Selecting the top-ranked backtest would have been equivalent to picking at random. The final basket was therefore chosen on structure and on how frequently a configuration family appeared across the surviving set, not on its score.

Several filters that looked convincing were tested and rejected on the same basis: a VIX threshold, an out-of-the-money strike shift, an opening-range ATR filter, and a green-candle break exit. Each improved the in-sample curve. None held up.

So when you are handed a NIFTY algo backtest, the questions that matter are: how many configurations were searched before this one was chosen, was the out-of-sample period genuinely untouched during development, and what happened to the candidates that were rejected. A single beautiful equity curve with no account of its siblings is not evidence.

09What building one involves

The research is roughly a third of the work. The rest is execution engineering, and it is where strategies that test well go to die.

• Strike and expiry selection against a real exchange calendar, including the weeks where a holiday shifts expiry.
• Broker-side stop-losses, so that a disconnect between your server and the broker does not leave a short option unprotected.
• Fill confirmation, because an order that was placed is not an order that was filled, and a strategy that assumes otherwise will silently run the wrong position size.
• Spike-trigger detection and order repricing, so that a momentary liquidity hole does not trigger an exit at an absurd price.
• A kill switch that flattens everything and stops trading on a defined condition, reachable without logging into a terminal.

None of this improves the backtest. All of it determines whether live results resemble it.

10Questions people ask