Why ask this question

Intraday option buying is popular because the payoff is asymmetric: a small premium can multiply on a strong move. It also needs direction to be right often enough, and quickly enough, to beat time decay and the bid-ask spread.

Our earlier research tested many directional entry signals on one-minute NIFTY data — moving-average and oscillator triggers, opening-range retests, Donchian breakouts and others. None survived. Testing signals one at a time has a weakness, though: a long list of failures does not prove the next idea will fail.

So this study asked a different question: what is the most that any linear, price-based signal could possibly predict about NIFTY's direction over the next few minutes? If that ceiling is below what an option buyer needs, there is no point searching further in that family.

The approach

The idea comes from signal processing, which is part of my electrical engineering background. A price series can be described by how its variation is distributed across frequencies. If returns contain repeating, predictable structure, it shows up as an uneven spectrum; if they behave like noise, the spectrum is flat. From the spectrum you can estimate an upper bound on how much of the next return's variance any linear filter of past returns could explain — without having to specify the filter.

We estimated this ceiling on one-minute NIFTY returns at horizons of 1, 5, 15 and 30 minutes — for all sessions, by hour of day and by India VIX level, each analysed separately (28 condition-by-horizon cells in total) — with the protocol and pass criteria pre-registered before the analysis was run.

We then converted the predictable share of variance into the best hit rate it could support, and compared that with the hit rate an at-the-money option buyer would need to break even after decay and costs at each horizon.

Results

Direction is almost unpredictable at the minute scale. The best-case linear predictability was about 0.2% of return variance at horizons of 1–5 minutes in the development period and at most about 0.7% in validation; at 15–30 minutes it was negligible in most periods.

That caps the hit rate near a coin flip. Translated into directional accuracy, predictability this low supports a best-case hit rate of roughly 50–52% across all sessions. An at-the-money option buyer needs somewhere between about 54% and 80%, depending on horizon and conditions, to cover decay and spread.

No condition rescued it. None of the 28 hour-of-day and VIX cells reached the required level.

Volatility, by contrast, is forecastable — and already priced. The size of the rest-of-day move, measured from 10:15, was predictable out of sample (R² of about 0.39 in the validation period and 0.65 in the 2026 holdout). But this did not help a straddle buyer. Typical rest-of-day moves of about 93–130 points (validation 2025 and January–May 2026) sat below the 123–146 points a straddle needed to break even, and straddle profit was unrelated to the volatility forecast. The options market had already priced what the forecast knew.

What this means

  • For linear, price-based direction signals, the search is closed. No moving average, oscillator or breakout rule built from past prices can beat a ceiling that sits near 50–52% across all sessions and that, in no condition tested, reaches what an option buyer needs. That explains, in one result, why the long list of individual triggers failed.
  • Forecasting volatility is not the same as profiting from it. Volatility is predictable, but option prices already reflect what is predictable. Being right about the size of the move is not enough if the premium paid already assumed it.
  • What the ceiling does not cover. The bound applies to linear functions of past prices. It does not rule out non-linear effects, information outside the price series, or rare events. The remaining leads we consider worth testing are of that kind: whether price jumps cluster in time (and whether an option position after a jump can profit from it), and whether order-flow imbalance in recorded bid-ask data carries short-term direction.

Limitations

  • One-minute bars; sub-minute structure is not covered.
  • The ceiling is estimated, with its own sampling error; small differences between horizons should not be over-read.
  • The breakeven hit rates depend on the cost and decay assumptions used and would shift for different strikes or expiries.

Conclusion

Rather than adding one more failed signal to the list, this study measured the ceiling that every such signal sits under. For one-minute NIFTY direction, that ceiling is close to chance, and well below what an intraday option buyer needs. The productive questions for option buyers lie elsewhere: in non-price information, non-linear behaviour, and the gap between what is forecastable and what is already priced.