Weekly, Monthly and All-Time-High Drawdowns on Nifty: the Widest Window Won 31 of 36

Lineage: Widens the single-day drawdown signal of the weakness study to a calendar week, a calendar month and the running all-time high.

The premise

Every study before this one measured weakness over a single day and found nothing usable. The obvious objection is that a day is too short a window to contain any information, and that a dip worth buying is a larger event measured over a longer span. This study tests that objection directly by holding the framework constant and varying only the window over which the fall is measured.

Three windows are tested. The weekly version, Strategy 5, asks whether a full calendar week closed below its own open by at least the threshold, buying at the following week's open. Its monthly counterpart, Strategy 8, asks the same of a calendar month. Third is Strategy 12, which asks whether Nifty has fallen by the threshold from its running all-time high, a condition satisfiable on any day rather than only at a period boundary.

A 5% weekly close-down and a 36% fall from an all-time high are not the same kind of event, and running all three on one framework measures what that difference is worth.

How the strategies work

Weekly and monthly drawdown share one formula: the period's open minus its close, as a fraction of the open, compared against the threshold, with the buy placed at the opening price of the following period. Both sweep 1.0% to 20.0% in 0.5% steps, 39 thresholds each.

All-time-high drawdown uses the running all-time high of the closing price minus the current close, as a fraction of that high. It is checked daily against the prior day's value and capped at one trade per month, so that a long bear market cannot deploy the same bank repeatedly within a single contribution cycle. It sweeps 5% to 40% in 1% steps, 36 thresholds.

Wider windows produce rarer and more extreme signals, and the rarity is the whole mechanism being tested.

The setup

A bank account receives ₹10,000 on the first trading day of each month, compounding daily at 5% annualised on actual calendar-day gaps. When a signal fires, the full available balance deploys after a 0.03% transaction cost. Units are held permanently, with no selling, no rebalancing and no capital gains tax, because no gain is ever realised.

The benchmark is a capital-matched monthly SIP on the same Nifty 50 daily data, no bank and no timing: 11.9907% XIRR over the full thirty-year period.

The best combinations

The weekly 11% row is flagged in the source study as an artefact rather than a replicable edge. It rests on four trades, and one threshold step in either direction destroys it. Weekly drawdown's genuine wins are 4.5%, 5% and 6.5%, four outperforming thresholds out of 39, and all three are small.

Monthly drawdown is the middle case in every respect: 18 of 39 thresholds outperform, clustered broadly between 2.5% and 11.5%, with the best at 7% adding ₹9.76 lakh across 38 trades.

All-time-high drawdown is the outlier. 31 of 36 thresholds outperform, and the winning region runs almost the entire sweep, from 8% through 38%. What matters is the width of that band rather than the peak figure. A winning region spanning thirty percentage points does not require the implementer to pick the right number, which is the objection that disqualified the strength signals in the previous study.

What the reference window does

The reference window drives almost everything here, and the threshold inside each window matters far less.

A 5% weekly close-down happens constantly on Nifty and usually reverses within days, so directing bank capital at it is a reaction to noise. A 36% fall from an all-time high is rare. Nifty spent most of the thirty-year window within 20% of its own peak, so a signal set at that level fires only during the 2001 to 2003 correction, the 2008 to 2009 crisis and the 2020 crash, and therefore selects for genuine distress rather than for ordinary variation.

The win rate roughly doubles with each widening: 4 of 39 for weekly, 18 of 39 for monthly, 31 of 36 for all-time-high. That progression is the finding. The threshold within a window is a second-order choice.

Weekly drawdown past 17% never fires at all across the tested window. The bank then compounds at 5% for thirty years while the index runs near 12%, and the shortfall against the SIP reaches ₹2.42 crore. That region is disqualified outright rather than badly tuned.

Where it fails

Each signal has a failure zone, and the zones are not alike.

Weekly drawdown fails through over-selectivity. Past 8.5%, additional wealth falls from −₹5.11 lakh to −₹27.20 lakh in a single step, and the floor is −₹2.42 crore once the signal stops firing altogether.

Monthly drawdown fails at the same end, reaching −₹84.09 lakh between 17.5% and 20%, driven by two trades. One of those is the 2020 crash, and the case is instructive: even a correctly timed entry at a genuine bottom does not offset years of capital sitting idle beforehand.

All-time-high drawdown fails at both ends, and narrowly. Between 5% and 7% the condition fires so often, 254 to 291 trades, that the strategy degenerates into an imperfect SIP. At 39% the bank holds back on 97% of days waiting for falls as deep as 2008 and 2020, leaving ₹31.72 lakh idle. Even that worst case, −₹11.98 lakh, is roughly one-fifth of monthly drawdown's worst case, because the strategy stays invested until the index falls hard from its peak and the downside is bounded by how rarely that happens.

One caution applies to the winner specifically. Deploying only in genuine bear markets requires holding cash through long stretches of ordinary years and then buying during the worst headlines of a decade, a behavioural demand this backtest does not model.

Verdict

Widening the reference point changed the answer, which is the first time anywhere in this series that a structural change to a dip-buying rule produced a usable result. All-time-high drawdown won 31 of 36 thresholds and added up to ₹30.99 lakh, monthly won 18 of 39 and topped out at ₹9.76 lakh, and weekly won 4 of 39 with its one spectacular figure resting on four trades. The asymmetry matters as much as the win rate: getting the all-time-high threshold wrong cost ₹11.98 lakh at worst, while getting the weekly threshold wrong cost ₹2.42 crore. On this evidence, the all-time high is the only reference point of the three worth using, and nothing narrower than it survives.

Appendix: assumptions and methodology

• Data. Nifty 50 daily open, high, low and close from nifty50_research_master.parquet, the full thirty-year period, index level only.
• Benchmark. Monthly SIP, ₹10,000 on the first trading day of each month, 0.03% cost, no bank and no signal: 11.9907% XIRR.
• Bank. ₹10,000 credited monthly, compounding daily at 5% annualised on actual calendar-day gaps until a signal triggers full deployment.
• Tax. None modelled, and none applies. Units are bought and held permanently, so no gain is ever realised.
• Signal definitions. Weekly and monthly drawdown fire when the period's open minus its close, divided by the open, reaches the threshold, buying at the following period's open. All-time-high drawdown fires when the running all-time high of the close minus the current close, divided by that high, reaches the threshold, checked against the prior day's value, at most one trade per month.
• Thresholds. Weekly and monthly, 1.0% to 20.0% in 0.5% steps, 39 levels each. All-time-high, 5% to 40% in 1% steps, 36 levels.
• Metrics. XIRR and additional wealth against the SIP. No Calmar or maximum-drawdown figure appears in these three workbooks' SUMMARY sheets.
• Not modelled. Year-by-year returns, split-period figures, brokerage beyond 0.03%, slippage, dividend reinvestment, and the behavioural cost of holding cash for years.
• Not tested. A Nifty 500 version of these signals, and whether the all-time-high band holds in the second half of the window when the 2008 and 2020 crashes are separated.
• Withheld. Nothing. The full sweep and the winning band are printed above, because the finding here is that the result does not depend on picking a precise threshold, and that claim cannot be made without showing the band it rests on.