A research note · published 3 November 2026 · data through 31 July 2026

Trend-Gated Diversification

A Supertrend breadth overlay for Indian broad-market indices

Ayush Agrawal · indexstrategy.ayushagrawal.in · educational document, not investment advice

1The question

The three Indian broad-market indices in this study move together. Their weekly returns correlate 0.82 to 0.95 with each other. So a basket of all three is really one bet — Indian equities. That bet earned 19.1% a year over the last five and a half years. But it fell -24.7% at its worst. For most people the fall, not the return, is what makes them sell at the wrong time.

So I asked one question: can a simple, mechanical, weekly rule keep most of the return while making the falls smaller? No forecasting. No fundamentals. Just a rule, written down in advance, that answers one thing every Friday: should the money be in the market this week, or not?

2The rule

The rule has two parts: a trend line on each of the three indices, and a gate that counts them.

Part one — a trend line per index. Every Friday, for each index, I draw one number: the Supertrend. It is a standard, textbook trend line (from Wilder’s 1978 book) built from the week’s price range and its average range over the last 10 weeks. An index above its line is in an uptrend; below it, a downtrend. That is the whole indicator. Nothing is tuned, nothing is learned from the data.

Part two — the gate. Count how many of the three indices are in an uptrend. That count, from 0 to 3, decides everything:

Look at the asymmetry. One index cracking is enough to leave. All three must be strong to return. The rule is quick to doubt and slow to trust. That asymmetry is not decoration; it is the result. The symmetric version — leave at two, return at two — whipsaws at the boundary (Section 7).

The holdings. When invested, the book holds 15% Nifty200 Momentum 30, 30% Nifty Midcap 150 and 55% Nifty Smallcap 250, with 10% kept in cash. The smallcap tilt is on purpose: smallcaps have the highest long-run return of the three, and once the gate is handling the risk, holding more of them costs little in extra fall (Section 8).

Mechanics. The signal is read on Friday’s close and traded at the next week’s open, so no trade ever uses the bar it is decided on. Costs of 0.25% per side are charged on every traded rupee. Cash earns 5% a year. No shorting, no leverage, nothing intra-week.

3The data

Every number in this paper comes from the official NSE bhavcopy archive — the daily index files at archives.nseindia.com/content/indices/ind_close_all_<DDMMYYYY>.csv — 2620 files, one per trading day from 1 January 2016 to 31 July 2026. No vendor API, no broker feed, no third-party index site, no ETF prices. Returns are index-level price returns (dividends not reinvested — a small, uniform understatement everywhere).

One data quirk, told plainly: the official archive publishes the Nifty200 Momentum 30 index only from 28 August 2020, and its intraday open/high/low only from January 2021. Before that, the daily open, high and low are set equal to the close. All returns are unaffected. The main window therefore starts 2021-01-15 — the first week all three indices trade, plus 20 weeks of warm-up for the indicator. This is a short window, and Section 9 says so.

The sample ends on 31 July 2026 for a regulatory reason, not a research one: SEBI’s rules for persons engaged solely in education forbid using market price data from the preceding three months.

4What happened

Figure 1 is the whole story on one page. The basket (grey, dashed) ends higher. The gated book (solid) ends a little lower, but it sits out the two ugliest stretches. Whether that trade was worth it depends on which stretches you care about — and on 2022, where the gate was wrong.

Figure 1. One rupee invested in each, weekly closes, log scale, 2021-01-15 to 2026-07-31. Cash earns 5%/yr; 0.25% a side on everything traded. The flat stretches of the solid line are weeks spent in cash. Labels mark the gate’s actual exits and re-entries.

Figure 2. Falls from the running high, same scale on both panels. The basket’s worst fall is -24.7%; the gated book’s is -17.0%.

Volatility is annualised. Sharpe and Sortino use monthly returns and a zero risk-free rate — Sharpe measures return per unit of wobble, Sortino return per unit of downside wobble. Calmar is annual return divided by the size of the worst fall. Source: NSE bhavcopy archive, index data through 31 July 2026.

Read Table 1 honestly. The gated book gives up -0.34% a month to the basket — about 15.2% versus 19.1% a year — and that gap is not statistically meaningful (t = -0.73). What it buys: a smaller worst fall, and a distribution with less crash-side skew (-0.52 to 0.17).

Figure 3. Calendar-year returns. The gated book loses the strong years and wins some weak ones — except 2022, where it lost more. That year is the paper’s honest centre.

The 2022 episode, in full. The gate exited on 4 March 2022 with one leg cracked, sat out the spring, and returned on 9 September after a strong rebound. Then it exited again on 30 December and stayed out until 2 June 2023. The result: the gated book fell -13.0% in 2022 while the passive basket fell -3.5%. This is what fast rebounds cost: the gate missed the recovery, twice. Any honest reading of this strategy has to carry that year.

The overlay is cheap to run: 9 trades in 5.5 years (about 1.6 a year), in cash 32% of the weeks. The full trade list is in the Appendix.

5The obvious objection

“All you did is hold less equity.” It is the right question. A constant basket-plus-cash mix is de-risked all the time; the gate is de-risked only when an index is in a weekly downtrend. Table 2 shows the difference that timing makes.

Same window, same costs, cash at 5%/yr. To get the gated book’s drawdown with a constant mix you must hold about 65% in the basket — which earns 12.3% a year, against the gated book’s 15.2%. And holding the full basket to match the gated book’s return buys you the 24.7% fall. The gate beats every constant mix at its own risk level.

6Why it works — and when it won’t

Trend breaks bunch up early in serious declines and persist through them. Recoveries strong enough to push all three indices back above their trend lines rarely reverse immediately. So the gate is insurance against long, grinding falls, and it pays the premium by being late into fast rebounds.

The same logic produces a list of situations where this rule will lose, stated plainly:

• Fast V-shaped recoveries. Re-entry needs all three legs. The rule is late by design — 2022 is the worked example.
• Choppy, range-bound years. Every two- or three-week swing can flip the gate and cost about half a percent in frictions each time.
• Strong, steady bull markets. The gate gives back a little return every year there is nothing to protect against. 2023 and 2024 in Figure 3 show exactly that.

7What failed first

This rule is the survivor of several dead ends. I list them because they explain why the rule looks the way it does.

• Timing a single index with a momentum oscillator. An earlier study tested a weekly oscillator (the Larry Williams Large Trade Index) on broad-market ETFs with profit targets of +130% to +300%. Over ten years it made 10.2% a year against 11.8% for simply holding, with the same -34% fall — two trades, years in cash, no edge. Lesson: entry timing has no measurable edge here; exit timing does.
• Selling on a red cross too. Adding “sell when the oscillator turns red” cut the fall but destroyed the return (5.7% vs 10.6% a year on the large-cap index) — fourteen whipsaw trades instead of two good ones. This is where the gate’s asymmetry came from.
• A daily version of the gate. The same rule on daily bars whipsawed (the return collapsed from 17% to 10% a year). Weekly bars are slow enough to hold and fast enough to matter.
• Volatility targeting and graduated exposure. Scaling the book to a fixed volatility target, or stepping exposure down 100/70/40/0 as the count fell, both cost return without beating the simple on/off gate. Simpler won.

8Breaking the rule on purpose

If a rule only works at its exact settings, it is fitted, not found. So I broke this one in the two ways it can be broken — parameters and weights — and report every cell, including the bad ones.

Nine combinations. Return stays between 13.5% and 16.7% a year and the worst fall between -9.1% and -21.0%. Nothing collapses. The defaults (10 × 3) sit in the middle of the grid, not at its best cell — which is the point of using defaults.

The results barely move when the weights move. The smallcap tilt is the only lever that changes the per-year figure, and it costs little in fall — the gate absorbs the extra volatility.

One more check, the honest kind. Everything was fixed on 2021-01-15 to 2023-12-31, then read once on 2024-01-01 to 2026-07-31. The gated book earned 24.2% with a -15.5% fall in the design half, and 4.8% with a -17.0% fall in the holdout half. The drawdown control carried over. The return did not — the holdout half was a weak period for this rule, and I say so rather than hide it. Finally, the search caveat: across the whole research programme behind this paper, roughly four hundred rule configurations were evaluated before this specification was frozen. That number is why the grids above are shown in full, and why every backtest should be read with a haircut.

9What I don’t trust about this result

• The window is short. Five and a half years, one market, one cycle. The mechanism — trend persistence — is documented across markets in the literature, but the specific numbers here are Indian, 2021–2026 numbers.
• 2022 cuts both ways. The gate’s worst failure and its best defence sit in the same year. Remove 2022 and the story changes; keep it and the story is honest.
• Price indices. Dividends (about 1.0–1.3% a year) are excluded everywhere.
• Execution at scale. 0.25% a side is realistic for retail size. Large money replicating the smallcap leg will pay more.
• Taxes. Gate exits inside a year are short-term capital gains in India; at 1.6 trades a year the drag is real and not deducted.
• Three-month cut. The last quarter of data is excluded by regulation, and with it whatever the indices did after July 2026.

10Closing

The rule in one sentence: hold the three broad Indian indices only while all three are in a weekly uptrend, and hold nothing while any one of them is not. On five and a half years of official NSE data that sentence earns 15.2% a year against the basket’s 19.1%, shrinks the worst fall from -24.7% to -17.0%, and whipsaws once, in 2022. It is not a return prediction. It is a rule about when to be present. Whether the smoother ride is worth the 4 points a year — statistically indistinguishable, but still 4 points — is a decision each reader can now make with all the numbers on the table.

If you find an error, I would genuinely like to hear about it: the reproduction details in the Appendix make every number in this paper checkable in an afternoon.

AAppendix — the 9 trades

Four exits and four re-entries in five and a half years. Read the 2022 rows slowly: the gate left on 4 March, came back 9 September, left again 30 December, and stayed out until June 2023. That sequence — not any table — is what this rule really is.

BReproduction

CReferences

1. Wilder, J. Welles. New Concepts in Technical Trading Systems. Trend Research, 1978. (ATR and the trend bands Supertrend is built from.)
2. Faber, Mebane T. “A Quantitative Approach to Tactical Asset Allocation.” The Journal of Wealth Management, 2007. (The modern template for one simple, mechanical timing rule.)
3. Moskowitz, Tobias J., Yao Hua Ooi, and Lasse Heje Pedersen. “Time Series Momentum.” Journal of Financial Economics, 2012. (Evidence that short-horizon trend persistence is real across markets.)
4. Harvey, Campbell R., Yan Liu, and Heqing Zhu. “…and the Cross-Section of Expected Returns.” Review of Financial Studies, 2016. (Why any backtest should be read with a multiple-testing haircut.)
5. Securities and Exchange Board of India. Circular on persons engaged solely in education, 29 January 2025; SECC (Fourth Amendment) Regulations, 2024. (The three-month data rule this document follows.)
6. National Stock Exchange of India. Daily index archive, bhavcopy section: archives.nseindia.com. (All data used here.)