---
title: "Trend and momentum: the evidence"
url: https://tipsheet.markets/trend/evidence/
description: "Every pre-registered test of trend and momentum on Indian markets: single trend rules after costs and tax, a 24-signal ensemble and timing luck, sector rotation and stock momentum. Null results reported as null."
updated: 2026-11-20
publisher: tipsheet (https://tipsheet.markets/)
cite: "tipsheet, “Trend and momentum: the evidence”, 20 November 2026, https://tipsheet.markets/trend/evidence/"
---

Markets[Trend & momentum](https://tipsheet.markets/trend/)

# What trend and momentum have been worth

Trend rules on Indian indices have cut the worst falls, but none has beaten buying and holding by more than luck would explain, and after tax they trailed on every large-cap index. Rotating into the strongest sectors has not paid either. The one result that holds up is stock momentum: last year's winning stocks have kept beating last year's losers. Everything here was written down before the results were seen.

Written for investors tempted by timing rules, and anyone who wants to see a null result stated plainly. Data to 20 Nov 2026.

## The record

Each question, what I found, and the number that decided it. The sections below show the workings.

| Question | Result | The deciding number |
| --- | --- | --- |
| Do single trend rules beat buying and holding, once every rule tried is counted? | Failed | Best deflated Sharpe ratio 0.15 across 35 rule-market pairs; 0.95 needed |
| After tax, do they keep up on large caps? | No | Behind by 1.3–5.2 points a year on every large-cap index |
| Is an ensemble of 24 signals a safer choice than picking one? | Passed | Worst five-year rank: 28th percentile against 6th for a typical single rule |
| Does the ensemble beat buying and holding? | Failed | Deflated Sharpe 0.01 with 84 trials |
| Does the trend score help forecast a fall or volatility? | Failed | No improvement out of sample |
| Does holding the top three sectors by past return beat holding all sectors? | Failed | 12-1: deflated Sharpe 0.30; 6-1: 0.65; blended (MO-H5): 0.56 |
| Has stock momentum paid in India? | Passed | Top fifth minus bottom fifth 1.9% a month, Newey-West t 5.4; survives a site-wide correction for multiple tests |
| Does momentum crash after the market has fallen for two years? | Failed | Points that way, misses the bar set in advance |
| A second review: do 9 more questions (volatility forecasts and scaling, the 52-week high, regime models, persistence, what followed big moves) pass the same bar? | None passed | 8 tests run, each registered in advance; volatility forecasts came closest |
| A third review: does momentum's own volatility, or the way winners rose, tell you anything; and do fast and slow trend signals behave differently? | 2 of 9 passed | Volatility-managed momentum and steady-rise momentum passed; no speed of trend beat another, or buying and holding, by more than luck |

## How the tests were run

The rules, assets, costs and tests were fixed in the pre-registered specification before any Indian result was computed, and every rule tried is counted. These are historical tests on indices, not signals.

Every rule pays 0.10% of value traded each way and fund running costs while invested. After-tax results apply Indian capital-gains rules by date and sell everything at the end. The deflated Sharpe ratio asks whether an edge over buy-and-hold is bigger than luck would produce across all the rules tried: below 0.95, it could be luck. [Read the specification](https://tipsheet.markets/research/trend_momentum_spec/).

## Trend following, full history

Each rule holds an asset only while its trend is up and sits in cash otherwise. The two grids give every rule on every asset (35 pairs) over its full history: the after-tax gap against buying and holding, and how much shallower or deeper the worst fall was. The full numbers are in the table below them.

### After tax, rule minus buy-and-hold

Percentage points a year, full history of each asset. Blue: the rule kept more after tax; red: buying and holding did.

After tax, rule minus buy-and-hold. Heat table, 7 rows by 5 columns. Values from −7.0 pts to +1.7 pts.

### Worst fall, rule minus buy-and-hold

Percentage points. Blue: the rule's worst fall was shallower than buying and holding's; red: deeper.

Worst fall, rule minus buy-and-hold. Heat table, 7 rows by 5 columns. Values from −2 pts to +39 pts.

Every rule on every asset in a table: returns, falls, time invested and the deflated Sharpe

Strategy minus buy-and-hold. Returns are annual (CAGR); drawdown is the worst peak-to-trough fall. After-tax uses Indian capital-gains rules by date.

*Strategy minus buy-and-hold. Returns are annual (CAGR); drawdown is the worst peak-to-trough fall. After-tax uses Indian capital-gains rules by date.*

| Asset | Rule | From | Rule CAGR | Buy-hold CAGR | After tax, rule minus buy-hold | Rule worst fall | Buy-hold worst fall | Time invested | Deflated Sharpe vs buy-hold | More |
| --- | --- | --- | --- | --- | --- | --- | --- | --- | --- | --- |
| Nifty 50 | 10-month average | 2000-04-04 | 10.8% | 12.5% | −1.7 pts | −44% | −60% | 70% | 0.01 |  |
| Nifty 50 | 200-day average | 2000-04-18 | 9.9% | 12.8% | −3.0 pts | −33% | −60% | 70% | 0.00 |  |
| Nifty 50 | 12-month momentum | 2000-07-04 | 9.7% | 12.7% | −3.0 pts | −40% | −60% | 64% | 0.00 |  |
| Nifty 50 | Blend of all four | 2000-07-04 | 11.0% | 12.7% | −2.7 pts | −26% | −60% | 59% | 0.01 |  |
| Nifty 50 | Breakout | 2000-07-03 | 9.3% | 12.8% | −3.9 pts | −31% | −60% | 55% | 0.00 |  |
| Nifty 500 | 10-month average | 1995-11-02 | 11.6% | 13.5% | −2.4 pts | −45% | −67% | 69% | 0.01 |  |
| Nifty 500 | 200-day average | 1995-11-10 | 12.2% | 13.6% | −1.6 pts | −39% | −67% | 68% | 0.01 |  |
| Nifty 500 | 12-month momentum | 1996-02-02 | 9.7% | 14.1% | −4.4 pts | −49% | −67% | 59% | 0.00 |  |
| Nifty 500 | Blend of all four | 1996-02-02 | 11.1% | 14.1% | −4.2 pts | −46% | −67% | 58% | 0.00 |  |
| Nifty 500 | Breakout | 1996-01-31 | 9.6% | 14.3% | −5.2 pts | −45% | −67% | 49% | 0.00 |  |
| Nifty Next 50 | 10-month average | 2003-09-02 | 16.2% | 17.3% | −1.3 pts | −37% | −72% | 74% | 0.03 |  |
| Nifty Next 50 | 200-day average | 2003-08-28 | 16.1% | 17.3% | −1.3 pts | −33% | −72% | 73% | 0.02 |  |
| Nifty Next 50 | 12-month momentum | 2003-12-02 | 12.0% | 16.3% | −4.3 pts | −44% | −72% | 66% | 0.00 |  |
| Nifty Next 50 | Blend of all four | 2003-12-02 | 14.2% | 16.3% | −3.2 pts | −35% | −72% | 62% | 0.01 |  |
| Nifty Next 50 | Breakout | 2003-11-10 | 12.3% | 16.6% | −4.4 pts | −41% | −72% | 63% | 0.00 |  |
| Nifty Midcap 150 | 10-month average | 2006-02-02 | 10.0% | 15.1% | −4.9 pts | −53% | −73% | 73% | 0.00 |  |
| Nifty Midcap 150 | 200-day average | 2006-01-18 | 16.2% | 15.2% | +0.8 pts | −37% | −73% | 72% | 0.11 |  |
| Nifty Midcap 150 | 12-month momentum | 2006-05-03 | 7.2% | 14.1% | −7.0 pts | −55% | −73% | 65% | 0.00 |  |
| Nifty Midcap 150 | Blend of all four | 2006-05-03 | 11.5% | 14.1% | −2.9 pts | −39% | −73% | 61% | 0.01 |  |
| Nifty Midcap 150 | Breakout | 2006-04-05 | 10.3% | 14.3% | −3.9 pts | −42% | −73% | 56% | 0.00 |  |
| Nifty Smallcap 250 | 10-month average | 2006-02-02 | 14.0% | 13.4% | +0.5 pts | −45% | −76% | 67% | 0.09 |  |
| Nifty Smallcap 250 | 200-day average | 2006-01-18 | 15.2% | 13.4% | +1.7 pts | −36% | −76% | 66% | 0.15 |  |
| Nifty Smallcap 250 | 12-month momentum | 2006-05-03 | 10.6% | 12.3% | −2.0 pts | −50% | −76% | 55% | 0.02 |  |
| Nifty Smallcap 250 | Blend of all four | 2006-05-03 | 13.3% | 12.3% | +0.3 pts | −37% | −76% | 58% | 0.09 |  |
| Nifty Smallcap 250 | Breakout | 2006-04-05 | 12.4% | 12.7% | −0.7 pts | −38% | −76% | 50% | 0.05 |  |
| Gold | 10-month average | 2001-06-04 | 11.9% | 14.0% | −3.0 pts | −23% | −26% | 78% | 0.00 |  |
| Gold | 200-day average | 2001-05-29 | 11.8% | 13.8% | −3.2 pts | −28% | −26% | 78% | 0.00 |  |
| Gold | 12-month momentum | 2001-09-04 | 12.3% | 14.0% | −2.4 pts | −26% | −26% | 72% | 0.00 |  |
| Gold | Blend of all four | 2001-09-04 | 11.1% | 14.0% | −4.9 pts | −22% | −26% | 63% | 0.00 |  |
| Gold | Breakout | 2001-08-09 | 12.3% | 14.0% | −2.8 pts | −23% | −26% | 73% | 0.00 |  |
| G-sec 5-year | 10-month average | 2002-07-02 | 7.3% | 7.4% | −0.1 pts | −5% | −7% | 94% | 0.04 |  |
| G-sec 5-year | 200-day average | 2002-06-24 | 7.0% | 7.4% | −0.6 pts | −6% | −7% | 94% | 0.00 |  |
| G-sec 5-year | 12-month momentum | 2002-10-03 | 6.8% | 7.3% | −0.9 pts | −5% | −7% | 62% | 0.01 |  |
| G-sec 5-year | Blend of all four | 2002-10-03 | 6.8% | 7.3% | −2.0 pts | −5% | −7% | 60% | 0.01 |  |
| G-sec 5-year | Breakout | 2002-09-05 | 7.0% | 7.3% | −0.3 pts | −6% | −7% | 93% | 0.01 |  |

Asset

### Gold: growth of 1 under each trend rule and buy-and-hold

Weekly, pre-tax, after costs.

**How to read it.** Log scale, so equal slopes are equal returns. Look first at the big falls, where the rule lines flatten out, then at where each line ends. Tax, shown in the table, widens the gap against the rules.

Buy and hold 200-day average 12-month momentum Blend

Loading chart…

*[Method](https://tipsheet.markets/methods/models/) [JSON](https://tipsheet.markets/data/v1/models/trend/curves/gold.0f596b88c7a9.json)*

## Sector momentum

Each month, hold the three NSE sector indices with the best 12-month (or 6-month) return, skipping the latest month, against an equal-weighted basket of all sectors. Only two rules were tried, so the bar for luck is lower; over the full period neither clears it.

| Rule | Window | Momentum CAGR | Equal-weight CAGR | Nifty 500 CAGR | Momentum minus EW | Months beating EW | Worst fall | Deflated Sharpe vs EW | More |
| --- | --- | --- | --- | --- | --- | --- | --- | --- | --- |
| mom12\_1 | full | 14.2% | 15.4% | 13.6% | −1.2 pts | 49% | −56% | 0.30 |  |
| mom12\_1 | first\_half | 16.2% | 17.1% | 14.4% | −0.9 pts | 54% | −56% | – |  |
| mom12\_1 | second\_half | 12.3% | 13.6% | 12.7% | −1.3 pts | 44% | −36% | – |  |
| mom12\_1 | since\_2012 | 13.7% | 16.1% | 14.2% | −2.4 pts | 47% | −36% | – |  |
| mom6\_1 | full | 16.8% | 15.4% | 13.6% | +1.4 pts | 50% | −53% | 0.65 |  |
| mom6\_1 | first\_half | 17.6% | 17.1% | 14.4% | +0.5 pts | 49% | −53% | – |  |
| mom6\_1 | second\_half | 16.2% | 13.6% | 12.7% | +2.6 pts | 51% | −42% | – |  |
| mom6\_1 | since\_2012 | 16.8% | 16.1% | 14.2% | +0.8 pts | 51% | −42% | – |  |

Rule

### Sector momentum against equal weight

Monthly, after costs.

**How to read it.** The heavier line is the momentum portfolio, the grey line the basket it chooses from. A gap that opens and closes is noise; the deflated Sharpe in the table says whether it is more than that.

Top-3 sector momentum Equal-weight sectors

Loading chart…

*[Method](https://tipsheet.markets/methods/models/) [JSON](https://tipsheet.markets/data/v1/models/momentum/curves/mom12_1.18ededf20b90.json)*

## Why an ensemble

Each of the 24 votes can be run alone as a rule: hold the Nifty 500 while it votes up, cash otherwise. Their results differ widely, from 9.1% to 16.5% a year, and which one wins depends on the period. The ensemble holds the share of votes that are up: 59% invested on average.

### Every rule: reward against the worst fall

Nifty 500, full history, after costs, before tax. Each dot is one of the 24 single rules; colour is the signal type.

**How to read it.** Up is better reward for risk; right is a shallower worst fall. Buying and holding sits far to the left. The single rules scatter widely. The ensemble ranks 9th of 25 on Sharpe without anyone having to know in advance which rule would do best.

Past return Price vs average Average crossover Channel position Ensemble Buy and hold

### Rank among the 25 strategies, rolling five-year windows

Percentile of each strategy's Sharpe ratio among all 25 in the five years to each date (100 = best).

**How to read it.** The best rule over the whole period spent 17% of five-year stretches in the bottom fifth; nobody holding it then would have known it would win. The ensemble spent 0% there. That is the case for it.

Ensemble Best single rule over the full period (Price vs average, 1 month) Worst single rule (Past return, 6 months)

Loading chart…

*Source: NSE total-return indices; gold in rupees (WGC); NSE 5-year G-sec index; 91-day T-bill (RBI); rupee per dollar from ECB reference rates. [Method](https://tipsheet.markets/methods/barometers/) [JSON](https://tipsheet.markets/data/v1/barometers/trend/ranks.baaafc9251bd.json)*

### Nifty 500: growth of 1

Weekly, after trading and fund costs, before tax. The signal is set at one close and acted on at the next.

**How to read it.** Log scale, so equal slopes are equal returns. The best and worst single rules are chosen with hindsight. The ensemble sits between them, with shallower falls than buying and holding.

Ensemble Buy and hold Best single: Price vs average, 1 month Worst single: Past return, 6 months

Loading chart…

*Source: NSE total-return indices; gold in rupees (WGC); NSE 5-year G-sec index; 91-day T-bill (RBI); rupee per dollar from ECB reference rates. [Method](https://tipsheet.markets/methods/barometers/) [JSON](https://tipsheet.markets/data/v1/barometers/trend/curves.8cf1238c4504.json)*

Every rule in a table

Every specification on the Nifty 500, full history, before tax.

*Every specification on the Nifty 500, full history, before tax.*

| Rule | Return a year | Volatility | Sharpe | Worst fall | Time invested | Turnover a year | More |
| --- | --- | --- | --- | --- | --- | --- | --- |
| Past return, 1 month | 14.1% | 14.3% | 0.54 | −28% | 58% | 19.4× |  |
| Price vs average, 1 month | 16.3% | 14.0% | 0.69 | −20% | 58% | 23.3× |  |
| Average crossover, 1 month | 14.4% | 14.3% | 0.56 | −26% | 58% | 13.6× |  |
| Channel position, 1 month | 16.3% | 14.2% | 0.68 | −21% | 60% | 21.7× |  |
| Past return, 2 months | 15.7% | 14.9% | 0.62 | −36% | 59% | 12.7× |  |
| Price vs average, 2 months | 15.3% | 14.6% | 0.60 | −31% | 60% | 14.5× |  |
| Average crossover, 2 months | 14.5% | 15.3% | 0.54 | −38% | 60% | 6.5× |  |
| Channel position, 2 months | 15.9% | 14.7% | 0.64 | −38% | 61% | 12.8× |  |
| Past return, 3 months | 14.6% | 14.8% | 0.56 | −42% | 59% | 10.3× |  |
| Price vs average, 3 months | 16.5% | 14.6% | 0.67 | −34% | 59% | 11.4× |  |
| Average crossover, 3 months | 13.9% | 15.7% | 0.49 | −44% | 60% | 4.3× |  |
| Channel position, 3 months | 16.0% | 14.6% | 0.64 | −40% | 61% | 11.1× |  |
| Past return, 6 months | 9.1% | 15.8% | 0.22 | −45% | 57% | 8.5× |  |
| Price vs average, 6 months | 14.7% | 14.9% | 0.56 | −40% | 59% | 7.9× |  |
| Average crossover, 6 months | 12.9% | 16.3% | 0.43 | −53% | 58% | 2.1× |  |
| Channel position, 6 months | 15.2% | 14.8% | 0.59 | −32% | 61% | 7.9× |  |
| Past return, 9 months | 9.6% | 15.4% | 0.25 | −52% | 58% | 7.0× |  |
| Price vs average, 9 months | 13.4% | 15.2% | 0.47 | −34% | 59% | 7.4× |  |
| Average crossover, 9 months | 9.5% | 16.1% | 0.24 | −49% | 59% | 1.6× |  |
| Channel position, 9 months | 12.7% | 14.5% | 0.45 | −38% | 61% | 7.6× |  |
| Past return, 12 months | 10.1% | 15.7% | 0.28 | −49% | 60% | 7.4× |  |
| Price vs average, 12 months | 11.4% | 15.0% | 0.36 | −38% | 59% | 8.0× |  |
| Average crossover, 12 months | 9.4% | 16.1% | 0.23 | −49% | 58% | 1.4× |  |
| Channel position, 12 months | 11.9% | 14.9% | 0.39 | −40% | 62% | 6.6× |  |
| Ensemble (all 24) | 13.9% | 13.0% | 0.57 | −33% | 59% | 8.9× |  |

Buying and holding: 14.1% a year, Sharpe 0.42, worst fall -67%. Short-horizon rules trade often; after Indian capital-gains tax their results would be markedly worse (see the single rules).

## Timing luck

Many rules are checked once a month. Which day? Here each rule is checked once every 21 sessions, on each of the 21 possible starting days. The bars show the range of annual returns that produced; the tick is the "tranched" version, which splits money across all 21 days, as Newfound Research recommends. The typical rule's range is 4.2 points a year.

Worst Range across rebalance dates tranched Best Spread

Past return, 1 month 8.3% 13.4% 5.1 pts

Price vs average, 1 month 4.3% 15.7% 11.4 pts

Average crossover, 1 month 6.6% 14.5% 7.9 pts

Channel position, 1 month 5.5% 15.9% 10.4 pts

Past return, 2 months 8.4% 13.9% 5.5 pts

Price vs average, 2 months 7.8% 14.6% 6.8 pts

Average crossover, 2 months 9.2% 12.7% 3.6 pts

Channel position, 2 months 8.9% 14.8% 5.8 pts

Past return, 3 months 10.9% 15.8% 5.0 pts

Price vs average, 3 months 9.2% 15.4% 6.2 pts

Average crossover, 3 months 9.4% 12.1% 2.8 pts

Channel position, 3 months 10.4% 15.5% 5.1 pts

Past return, 6 months 7.2% 12.0% 4.9 pts

Price vs average, 6 months 10.4% 13.9% 3.5 pts

Average crossover, 6 months 8.9% 12.2% 3.3 pts

Channel position, 6 months 10.1% 14.3% 4.2 pts

Past return, 9 months 8.0% 11.7% 3.7 pts

Price vs average, 9 months 9.4% 12.4% 3.0 pts

Average crossover, 9 months 9.1% 10.9% 1.8 pts

Channel position, 9 months 9.2% 13.5% 4.2 pts

Past return, 12 months 8.2% 10.2% 2.1 pts

Price vs average, 12 months 8.8% 11.7% 2.9 pts

Average crossover, 12 months 9.4% 11.7% 2.2 pts

Channel position, 12 months 8.6% 12.3% 3.6 pts

Ensemble (all 24) 9.9% 13.2% 3.2 pts

Annual return of each rule on the Nifty 500 checked every 21 sessions, by starting day, full history, before tax. Tranching does not raise the expected return; it removes the luck of the draw.

## What happened next

Nifty 500 total returns after every session in each score band. A trend score describes the past; this figure shows how little it has said about the next few months on its own.

### After each score band: the next 3 and 12 months

Nifty 500 total return after every session in the band, full history. Today's band, 60–80, is in bold.

**How to read it.** Each dot is the median Nifty 500 return after sessions in that band, and the bar is the middle half of those returns. The shaded column is all sessions. The number on the right counts separate spells that do not overlap: thousands of sessions are only this many independent pieces of evidence. A band whose bar sits over the shaded column has said little; hollow rows have under five spells.

After each score band: the next 3 and 12 months. 2 horizons, 5 bands. 0–20: median +4.3%, 52 spells; 20–40: median +3.5%, 64 spells; 40–60: median +2.0%, 76 spells; 60–80: median +3.6%, 77 spells; 80–100: median +4.9%, 77 spells. All sessions: median +4.0%.

*[Method](https://tipsheet.markets/methods/barometers/)*

Every column in a table: share up and the worst fall too

Nifty 500 total return after each session in the band, full history.

*Nifty 500 total return after each session in the band, full history.*

| Reading | Sessions | Separate 3-month spells | Next 3 months, median | Next 3 months, middle half | 3 months: share up | Worst fall within 3 months, median | Next 12 months, median | More |
| --- | --- | --- | --- | --- | --- | --- | --- | --- |
| 0–20 | 1,528 | 52 | +4.3% | −4% to +13% | 64% | −5.3% | +14.1% |  |
| 20–40 | 931 | 64 | +3.5% | −5% to +11% | 61% | −5.3% | +9.7% |  |
| 40–60 | 1,064 | 76 | +2.0% | −6% to +9% | 56% | −5.7% | +7.8% |  |
| 60–80 | 1,029 | 77 | +3.6% | −4% to +10% | 64% | −3.7% | +11.5% |  |
| 80–100 | 3,033 | 77 | +4.9% | −3% to +12% | 66% | −3.7% | +13.7% |  |
| All sessions | 7,585 | 121 | +4.0% | −4% to +11% | 63% | −4.4% | +12.0% |  |

## The tests

Written down before any result was computed. The question was never whether trend following beats the market here; an earlier test said no. It was whether an ensemble removes the fragility of single rules, and what it costs.

1. TR-H1 Is the ensemble a safer choice than picking one specification? Passed Yes. Across 310 rolling five-year windows its worst rank among the 25 strategies was the 28th percentile, against the 6th for the typical single specification. Its Sharpe ratio over the full period, 0.57, was above the typical single's 0.54.
   
   Worst rank in any of 310 five-year windows (percentile among 25) typical single rule 6th **28th**
2. TR-H2 How much does the day of the month you check a rule matter? Measured A lot. Checked once every 21 sessions, the typical specification's annual return moved by 4.2 points depending on the day; the worst case was 11.4 points. The ensemble's spread was 3.2 points.
3. TR-H3 Does the ensemble beat buying and holding, once the number of rules tried is counted? Failed No. 13.9% a year against 14.1%, with a far smaller worst fall (-33% against -67%). The deflated Sharpe ratio is 0.01 with 84 trials counted; it would need 0.95. Before tax.
   
   Deflated Sharpe ratio against buying and holding, 84 trials needed 0.95 **0.01**
4. TR-H4 Does the score add to a forecast of volatility or a 10% fall? Failed No. For a 10% fall within three months p = 0.78; for volatility p = 0.03, but with opposite signs in the two halves of the sample, and neither improved out-of-sample forecasts.
   
   p-value, score added to the odds of a 10% fall bar 0.025 1 0.1 0.01 0.001 0.0001 **0.78**
   
   p-value, score added to the volatility forecast bar 0.025 1 0.1 0.01 0.001 0.0001 **0.035**
   
   p-values on a log scale; smaller is stronger evidence, to the right. Shaded: the side that passes the bar set in advance.
   
   - 10% fall: same sign in both halves: met
   - Volatility: same sign in both halves: not met
   - Improves out-of-sample forecasts: not met

Tests, samples and pass bars were fixed in [the pre-registered specification](https://tipsheet.markets/research/trend_barometer_spec/) before any result was computed. A pass needs every condition written there, not only a small p-value.

## How it is built

Four signal types at six horizons, on the excess return over cash. Equal weights, nothing fitted.

**The signals.** Past return (is the price above where it was h sessions ago?), price against its average (above its h-session average?), average crossover (is a quarter-length average above the h-session one?) and channel position (is the price in the upper half of its h-session range?). Horizons are 21, 42, 63, 126, 189 and 252 sessions. Each "up" is one vote; the score is the share of up votes.

**Why these.** Levine and Pedersen (2016) show that past-return and moving-average rules are close relatives, both weighted sums of past returns. So diversifying across horizons matters more than across signal types. Corey Hoffstein and colleagues at Newfound Research have written at length on two other risks of single rules. One is specification risk: the best lookback in a backtest is close to random out of sample. The other is timing luck (see "Rebalance Timing Luck: The (Dumb) Luck of Smart Beta", 2020). Their remedy for both is an ensemble with tranched rebalancing, which is what this page measures. I follow the method, not any product.

**Trend environment.** Niels Kaastrup-Larsen's Trend Barometer (Top Traders Unplugged) reads how strongly markets are trending. Its method is not public, so mine is a separate, disclosed measure in the same spirit.

**Execution and costs.** Signals at one day's close, trades at the next day's close. 0.10% of value traded each way; index-fund running costs while invested; the cash leg earns the 91-day T-bill rate less a liquid fund's costs.

## Stock momentum: is it paying?

Three readings, each against its own history. Momentum's return: the top fifth of stocks by the blend minus the bottom fifth, over the last 12 months (gross, before costs). Momentum breadth: the share of the 1,363 liquid stocks whose 12-1 month return is positive. The crash flag: whether the Nifty 500's total return over two years is negative.

**Momentum's return, 12 months**

+24% 46th percentile, to Oct 2026

**Momentum breadth**

40% of stocks up over 12-1 months; 18th percentile

**Crash flag**

Off the market is above where it was two years ago

### Momentum's 12-month return

Top fifth minus bottom fifth by the blended score, rolling 12 months, gross. Monthly.

**How to read it.** Above zero, past winners kept winning. The deep plunges below zero are momentum crashes; note when they came.

Loading chart…

*Source: NSE bhavcopy (main board), prices adjusted for corporate actions; liquid point-in-time universe; NSE Indices for Nifty200 Momentum 30. [Method](https://tipsheet.markets/methods/barometers/) [JSON](https://tipsheet.markets/data/v1/barometers/momentum/readings.34491b4b85ce.json)*

## Momentum's own risk, and how its winners rose

Two readings added in October 2026, after a second round of tests. Momentum's risk comes and goes, and it can be measured from the factor's own recent swings: scaling the long-short down when those swings are large and up when they are calm raised its Sharpe ratio from 1.02 to 1.43 and halved its worst month, a test I set before running it. And winners that rose through many small gains kept winning more than those that rose in a few jumps.

**Momentum's volatility, last six months**

10% a year; 35th percentile of its history, at Oct 2026

**Exposure for a 12% target**

1.19×the long-short scaled to 12% volatility; above 1 when momentum is calm

**Many small gains against a few jumps**

−9% continuous minus discrete momentum, last 12 months; 16th percentile

### Momentum, plain and volatility-managed: growth of 1

Top fifth minus bottom fifth by the blend, monthly, gross, log scale. The managed line holds 12% ÷ the factor's volatility over the last 126 sessions (weights 0.4× to 1.9×), from Feb 2005, drawn at the plain line's overall volatility so the two compare fairly.

**How to read it.** The managed line steps back before the worst crashes, because the factor's swings had already grown. At the plain factor's volatility, its worst month was −24% against −45% (May 2009). It is a measure of the factor, with leverage that changes monthly and no costs; not a fund anyone can buy.

Scaled by its own volatility Plain long-short

Loading chart…

*Source: NSE bhavcopy (main board), prices adjusted for corporate actions; liquid point-in-time universe. [Method](https://tipsheet.markets/methods/barometers/) [JSON](https://tipsheet.markets/data/v1/barometers/momentum/v3_series.5b68404b9dc2.json)*

### Momentum among stocks that rose steadily, and among those that jumped

Winners minus losers by 12-1 month return, within the third of each group whose year came from the most small moves (continuous) or the fewest (discrete). Monthly, gross, log scale.

**How to read it.** Continuous momentum averaged +2.22% a month and discrete +1.34%. The gap passed the bar I set by a hair (p 0.0121 against 0.0125), and it was much smaller in the second half of the history than the first.

Continuous: many small moves Discrete: a few large moves

Loading chart…

*Source: NSE bhavcopy (main board), prices adjusted for corporate actions; liquid point-in-time universe. [Method](https://tipsheet.markets/methods/barometers/) [JSON](https://tipsheet.markets/data/v1/barometers/momentum/v3_series.5b68404b9dc2.json)*

## Rotating into the leading sectors

Each month, hold the three of 13 non-overlapping sectors with the highest blended score, against all 13 equally weighted, after costs. This was written down in advance, as the third sector-momentum rule I have tried. It failed: 15.9% a year against 14.5%, beating the basket in 49% of months; the edge was positive in the first half and negative in the second; deflated Sharpe 0.56 against the 0.95 needed. Formed at the end of Oct 2026, it would hold Pharma, Media, Metal.

### Sector momentum against all sectors

Growth of 1, monthly, after costs; 13 non-overlapping NSE sector indices, total returns.

**How to read it.** The heavier line rotates into the three strongest sectors each month; the grey line holds them all. A gap that opens and closes again is noise.

Top three sectors by blended momentum All sectors, equal weight

Loading chart…

*Source: NSE Indices total-return series (dividends reinvested). [Method](https://tipsheet.markets/methods/barometers/) [JSON](https://tipsheet.markets/data/v1/barometers/momentum/sector_curves.d416fefdd75d.json)*

## Six flavours, blended

Momentum can be measured many ways. The blend z-scores each of the six within the universe every month, caps extremes at three standard deviations, and averages them with equal weights. Below: each flavour's long-short record since Dec 2003, and how alike their rankings are.

### Top fifth minus bottom fifth: growth of 1

Monthly, gross of costs, log scale. A factor measure, not an investable portfolio.

**How to read it.** Log scale. Look at the steep drops, which are momentum crashes. Residual momentum, which strips out the market's own moves, has the steadiest line.

Blend 12-1 month Residual 52-week high

Loading chart…

*Source: NSE bhavcopy (main board), prices adjusted for corporate actions; liquid point-in-time universe; NSE Indices for Nifty200 Momentum 30. [Method](https://tipsheet.markets/methods/barometers/) [JSON](https://tipsheet.markets/data/v1/barometers/momentum/ls_curves.295669f61543.json)*

### Each flavour: reward against the worst fall

Long-short (top fifth minus bottom fifth), monthly, gross, since the start of the sample.

**How to read it.** Up is a better Sharpe ratio; right is a shallower worst fall. The blend is the large dot.

Single flavour Blend of all six

### How alike the flavours are

Average month-by-month rank correlation between flavours. The stronger the colour, the more alike; faint pairs bring the most variety to a blend.

|  | 12-1 | 6-1 | Vol-adj. | Residual | 52-wk high |
| --- | --- | --- | --- | --- | --- |
| 6-1 | 0.67 |  |  |  |  |
| Vol-adj. | 0.97 | 0.66 |  |  |  |
| Residual | 0.73 | 0.53 | 0.74 |  |  |
| 52-wk high | 0.52 | 0.53 | 0.54 | 0.35 |  |
| NSE score | 0.82 | 0.82 | 0.85 | 0.64 | 0.71 |

The flavours in tables: long-short and long-only

Long-short (top fifth minus bottom fifth), monthly, gross.

*Long-short (top fifth minus bottom fifth), monthly, gross.*

| Flavour | Average month | Volatility | Sharpe | Worst fall | More |
| --- | --- | --- | --- | --- | --- |
| Residual (stock-specific) 12-1 | 1.38% | 14.0% | 1.19 | −39% |  |
| NSE's normalised score | 1.90% | 20.3% | 1.12 | −49% |  |
| Blend of all six | 1.94% | 21.6% | 1.08 | −55% |  |
| Volatility-adjusted 12-1 | 1.67% | 20.5% | 0.98 | −61% |  |
| 6-1 month return | 1.53% | 18.8% | 0.98 | −39% |  |
| 12-1 month return | 1.61% | 23.1% | 0.84 | −68% |  |
| Nearness to 52-week high | 1.57% | 27.2% | 0.69 | −66% |  |

Average month-by-month rank correlation between flavours. Lower means more variety to blend.

*Average month-by-month rank correlation between flavours. Lower means more variety to blend.*

|  | 12-1 | 6-1 | Vol-adj. | Residual | 52-wk high | NSE score | More |
| --- | --- | --- | --- | --- | --- | --- | --- |
| 12-1 month return | 1.00 | 0.67 | 0.97 | 0.73 | 0.52 | 0.82 |  |
| 6-1 month return | 0.67 | 1.00 | 0.66 | 0.53 | 0.53 | 0.82 |  |
| Volatility-adjusted 12-1 | 0.97 | 0.66 | 1.00 | 0.74 | 0.54 | 0.85 |  |
| Residual (stock-specific) 12-1 | 0.73 | 0.53 | 0.74 | 1.00 | 0.35 | 0.64 |  |
| Nearness to 52-week high | 0.52 | 0.53 | 0.54 | 0.35 | 1.00 | 0.71 |  |
| NSE's normalised score | 0.82 | 0.82 | 0.85 | 0.64 | 0.71 | 1.00 |  |

Long-only: hold the top fifth, equal weight, after 0.25% trading cost each way, against all liquid stocks equally weighted. Price returns.

*Long-only: hold the top fifth, equal weight, after 0.25% trading cost each way, against all liquid stocks equally weighted. Price returns.*

| Flavour | Return a year | All stocks | Difference | Worst fall | Turnover a year | More |
| --- | --- | --- | --- | --- | --- | --- |
| 12-1 month return | 20.6% | 12.2% | +8.3 pts | −74% | 2.9× |  |
| 6-1 month return | 19.3% | 12.2% | +7.0 pts | −71% | 4.3× |  |
| Volatility-adjusted 12-1 | 22.8% | 12.2% | +10.5 pts | −73% | 3.0× |  |
| Residual (stock-specific) 12-1 | 19.1% | 12.2% | +6.8 pts | −74% | 3.3× |  |
| Nearness to 52-week high | 19.9% | 12.2% | +7.6 pts | −68% | 5.2× |  |
| NSE's normalised score | 25.2% | 12.2% | +13.0 pts | −72% | 3.2× |  |
| Blend of all six | 25.1% | 12.2% | +12.8 pts | −72% | 2.9× |  |

## The crash flag

Daniel and Moskowitz (2016) found that momentum's crashes come in "panic states": after a long market fall, when the market rebounds, last year's losers (the most beaten-down stocks) jump and last year's winners lag. My flag is their bear-market indicator: the Nifty 500's two-year return below zero.

### Every month of momentum, with the crash flag

Long-short return of the blend each month, gross. Shaded: months when the flag was on going in.

**How to read it.** Momentum gained in 72% of months. The deep red bars are the crashes; see how many sit in the shaded spans, which follow a two-year fall in the market.

Momentum lost Momentum gained Crash flag on

The ten worst months, in a table

The ten worst months for the blend's long-short, and whether the flag was on going into them.

*The ten worst months for the blend's long-short, and whether the flag was on going into them.*

| Month | Long-short return | Flag |
| --- | --- | --- |
| May 2009 | −45.0% | On |
| Jan 2012 | −23.1% | On |
| Jun 2020 | −20.2% | On |
| Dec 2008 | −17.6% | On |
| Oct 2013 | −13.1% | Off |
| Apr 2026 | −11.7% | Off |
| Aug 2020 | −11.0% | On |
| May 2014 | −10.9% | Off |
| Mar 2014 | −10.7% | Off |
| Apr 2020 | −10.6% | On |

The flag was on in 30 months since 2003. In those months the long-short averaged −1.49%; in the rest, +2.38%. The difference narrowly misses the significance bar I set in advance.

## Timing luck in Nifty200 Momentum 30

NSE's Nifty200 Momentum 30 picks 30 stocks twice a year, in June and December. I rebuilt it from NSE's published method, then ran the same rules on the five other half-yearly calendars. Same stocks, same rules, different months: annual returns ranged from 13.5% to 21.8% since May 2005, a spread of 8.4 points a year. In 2009 the best and worst calendars differed by 60 points.

The replica on each half-yearly calendar, price returns, before costs.

*The replica on each half-yearly calendar, price returns, before costs.*

| Rebalanced in | Return a year | Volatility | Sharpe | Worst fall | More |
| --- | --- | --- | --- | --- | --- |
| Jan and Jul | 16.7% | 27.0% | 0.72 | −72% |  |
| Feb and Aug | 20.6% | 27.9% | 0.82 | −72% |  |
| Mar and Sep | 21.8% | 28.0% | 0.85 | −72% |  |
| Apr and Oct | 20.9% | 26.9% | 0.85 | −71% |  |
| May and Nov | 17.4% | 26.6% | 0.75 | −76% |  |
| Jun and Dec (NSE's) | 13.5% | 26.8% | 0.62 | −78% |  |

### Each year, how far apart the six calendars ended

Calendar-year return of the replica on each half-yearly rebalance calendar, price returns before costs. Full years only.

**How to read it.** Each row is a year and each dot one calendar; the line spans the best and worst. The longer the line, the more the choice of months mattered that year. NSE's own calendar is the bold dot.

The other five calendars June and December (NSE's)

Each year, how far apart the six calendars ended. Dot plot of 20 items. Jan and Jul: highest 2007 at 102%, lowest 2008 at −69%.

The same returns as a table

*Calendar-year returns of the replica by rebalance calendar (months), %. Full years only.*

|  | Jan, Jul | Feb, Aug | Mar, Sep | Apr, Oct | May, Nov | Jun, Dec |
| --- | --- | --- | --- | --- | --- | --- |
| 2006 | +37 | +52 | +52 | +45 | +28 | +39 |
| 2007 | +102 | +131 | +113 | +97 | +97 | +85 |
| 2008 | −69 | −69 | −69 | −68 | −74 | −75 |
| 2009 | +61 | +121 | +106 | +114 | +72 | +62 |
| 2010 | +12 | +13 | +10 | +4 | +4 | +13 |
| 2011 | −25 | −31 | −23 | −18 | −20 | −26 |
| 2012 | +44 | +43 | +38 | +41 | +43 | +43 |
| 2013 | −9 | −3 | 0 | +8 | −4 | −6 |
| 2014 | +68 | +83 | +88 | +85 | +70 | +46 |
| 2015 | +7 | +16 | +8 | +10 | 0 | +2 |
| 2016 | −5 | −5 | +5 | +1 | +5 | 0 |
| 2017 | +55 | +62 | +65 | +73 | +77 | +55 |
| 2018 | −19 | −15 | −14 | −18 | −15 | −8 |
| 2019 | +11 | +9 | +10 | +5 | +3 | +3 |
| 2020 | +26 | +35 | +46 | +34 | +42 | +41 |
| 2021 | +68 | +47 | +57 | +58 | +69 | +65 |
| 2022 | −8 | −3 | −4 | −6 | +1 | −6 |
| 2023 | +45 | +56 | +66 | +61 | +64 | +52 |
| 2024 | +30 | +35 | +34 | +39 | +43 | +23 |
| 2025 | +10 | −2 | +1 | +8 | +2 | −10 |

### The replica: NSE's calendar against all six combined

Growth of 1, monthly, log scale, price returns before costs.

**How to read it.** The heavier line is the replica on NSE's own months; the grey line splits money across all six calendars, which removes the luck of the draw without changing the expected return.

June and December (NSE's calendar) Split across all six calendars

Loading chart…

*Source: NSE bhavcopy (main board), prices adjusted for corporate actions; liquid point-in-time universe; NSE Indices for Nifty200 Momentum 30. [Method](https://tipsheet.markets/methods/barometers/) [JSON](https://tipsheet.markets/data/v1/barometers/momentum/replica.27f1e199130f.json)*

**How close is the replica?** Its monthly returns on NSE's calendar correlate 0.91 with the real index over 258 months, above the 0.90 I required before naming the index. Its level is lower, though. It uses price returns without dividends. I have no free-float market values, so weights follow the momentum score alone, with NSE's 5% cap. Membership of the Nifty 200 is approximated by the 200 most-traded stocks. The spread between calendars is the finding; the exact returns are the replica's, not the index's. That NSE's calendar came out lowest is itself luck.

## The tests

Written down before any result was computed.

1. MO-H1 Has stock momentum paid in India? Passed Yes. Stocks in the top fifth of the blend beat the bottom fifth by 1.9% a month on average since Dec 2003 (Newey-West t 5.4), in both halves of the sample. Held alone and after trading costs, the top fifth returned 25.1% a year against 12.2% for an equal-weighted basket of every liquid stock; deflated Sharpe 0.99 with 16 trials.
   
   p-value, top fifth minus bottom fifth (Newey-West) bar 0.017 1 0.1 0.01 0.001 0.0001 **below 0.0001**
   
   p-values on a log scale; smaller is stronger evidence, to the right. Shaded: the side that passes the bar set in advance.
   
   - Positive in the first half: met
   - Positive in the second half: met
2. MO-H2 Does blending the flavours beat the typical single flavour? Failed Half. The blend's Sharpe ratio (1.08) beat the typical flavour's (0.98), but its worst fall (-55%) was deeper than the typical flavour's (-55%), so it fails the bar set in advance. Residual momentum on its own had the best Sharpe and the shallowest fall.
   
   Sharpe ratio of the blend typical flavour 0.98 **1.08**
   
   Worst fall of the blend, % typical flavour −55% **−55%**
3. MO-H3 Does momentum crash after the market has fallen for two years? Failed It points that way but misses the bar. In the 30 months after the Nifty 500's two-year return turned negative, the long-short return averaged −1.49% a month, against +2.38% otherwise (p = 0.04; the bar was 0.017). Of the ten worst months, 6 were flagged, including the worst four.
   
   p-value, flagged months against the rest (Welch) bar 0.017 1 0.1 0.01 0.001 0.0001 **0.044**
   
   p-values on a log scale; smaller is stronger evidence, to the right. Shaded: the side that passes the bar set in advance.
4. MO-H5 Does rotating into the sectors with the strongest blended momentum beat holding them all? Failed No. 15.9% a year against 14.5% for an equal-weighted basket of 13 non-overlapping sectors since Feb 2005, after costs; ahead in 49% of months; deflated Sharpe 0.56 with 3 sector rules counted.
   
   Deflated Sharpe ratio against the equal-weight basket, 3 trials needed 0.95 **0.56**
   
   - Ahead in the first half: met
   - Ahead in the second half: not met
5. MO-H4 Which flavours bring the most variety? Measured The least correlated pair is residual (stock-specific) 12-1 and nearness to 52-week high (0.35), as the literature suggests. Volatility-adjusted and plain 12-1 momentum are near-duplicates (0.97).

Tests, samples and pass bars were fixed in [the pre-registered specification](https://tipsheet.markets/research/momentum_barometer_spec/) before any result was computed. A pass needs every condition written there, not only a small p-value.

## How it is built

Every month-end, every liquid NSE main-board stock, using only data up to that day.

**Universe.** The same point-in-time universe as my breadth measures: main-board stocks with a median traded value of at least ₹1 crore a day (in today's rupees) and trades on 45 of the last 60 sessions, with at least 273 sessions of history. Stocks that later delisted are included while they traded, so there is no survivorship bias. Prices are bhavcopy closes adjusted for splits, bonuses and other corporate actions; dividends are not added.

**The six flavours.** 12-1 and 6-1 month returns (Jegadeesh and Titman, 1993); 12-1 divided by volatility (in the spirit of Barroso and Santa-Clara, 2015); residual momentum, the stock's own return after taking out the market's, over the same months (Blitz, Huij and Martens, 2011); nearness to the 52-week high (George and Hwang, 2004); and NSE's own normalised score. Industry momentum needs a point-in-time industry map, which I do not have. Earnings momentum waits for company financials.

**Portfolios.** Fifths by score, equally weighted, held one month. A stock that stops trading earns nothing after its last trade. The long-only version pays 0.25% each way on what it trades.

## The rules across every market

The 10-month and 200-day rules were tested on seven markets. Here they are replayed on all 127 investable markets in the section, each over its own full history, after costs and tax. After tax the 10-month rule kept more than buying and holding on 32 of 127 and the 200-day rule on 24; the typical gap was -1.7 and -1.8 points a year. Their worst falls were shallower on 126 and 124. This is a description, not a test: the markets overlap, many histories are back-calculated, and with this many tries some markets would favour the rules by luck alone.

### After tax, the 10-month rule mostly trailed buying and holding

Rule minus buy and hold, percentage points a year after tax, full history of each market. One dot per market; colour is the family.

**How to read it.** Right of zero the rule kept more after tax. Each family's best and worst are named; hover any dot for its name.

### After tax, the 200-day rule mostly trailed buying and holding too

Rule minus buy and hold, points a year after tax.

### What the rules did buy: shallower worst falls

The 200-day rule's worst fall minus buying and holding's, percentage points. Right of zero: shallower.

**How to read it.** Almost every dot is right of zero. This is what trend rules have reliably offered: a smaller worst fall, paid for with lower returns after tax.

## The second review: ideas from the research, tested

In October 2026 I went back to the research on trend, momentum and risk and picked the ideas with the strongest record: volatility forecasts, volatility-scaled trend, the 52-week high, fast and slow trend blends, persistence tests, regime-switching models, and what followed past trends; then, at a reader's suggestion, what followed 200-day crossings, big falls and new highs. I wrote down each test and its pass mark before running it. None passed. The closest was the volatility forecast, which beat trailing volatility on nearly every market but not steadily enough to meet the bar I had set. So the site adds no new signal, and the trend score and regimes stay what they were: descriptions of the past.

Each test was registered before its result was seen, with a sample from each index's launch onward, out-of-sample checks, a bootstrap that keeps markets' joint moves, and a correction for the number of variants tried. The full specification and log are linked below.

*Each test was registered before its result was seen, with a sample from each index's launch onward, out-of-sample checks, a bootstrap that keeps markets' joint moves, and a correction for the number of variants tried. The full specification and log are linked below.*

| Question | Result | The deciding number |
| --- | --- | --- |
| Can a volatility forecast beat last quarter's volatility as a guide to next month's? | Near miss | GARCH better on 97% of 126 markets and EWMA on 94%, but neither held up in both halves of the history, which I required in advance |
| Does a calibrated range for the next month hold its promised coverage? | Not run | Built only on a forecast that passes the first test |
| Does scaling trend rules by volatility make them better? | Failed | It made them worse: the ensemble's Sharpe ratio fell by 0.07 (−0.12 to −0.01), lower on 91% of markets |
| Do sectors near their 52-week high beat those far from it? | Failed | +0.12% a week (t 1.5), and close to zero in the second half |
| Does blending fast and slow trend signals help? | Failed | Against holding: Sharpe −0.07 (−0.33 to +0.21); against the slow signal alone +0.10 (−0.05 to +0.24) |
| Do index returns persist from week to week? | No reliable persistence | Significant on 4% of 107 markets at four weeks and 1% at thirteen; a quarter needed |
| Does a Markov regime model beat the simple 30/70 band? | Failed | Sharpe +0.01 against the band (−0.23 to +0.24), ahead on 51% of markets |
| After an uptrend begins, do returns beat those after a downtrend begins? | Failed | 2,406 trends: about zero over 26 weeks (−3.3 to +3.7), +4.0 points over 52 (−2.7 to +10.8) |
| After an index falls below its 200-day average, falls 10% or 20%, or makes a new 52-week high, do the next year's returns or falls differ from usual? | Failed | 106 markets, ten comparisons: none survives the correction for ten tries. Closest: a year after a 52-week high, +5.8 points over cash (p 0.03); after a fall below the 200-day average, the next year's deepest fall −2.2 points (p 0.03) |

### Volatility forecasts beat last quarter's volatility almost everywhere

How much better a GARCH(1,1) forecast of the next month's volatility was than the past 63 sessions' volatility, out of sample from 2001, by the QLIKE loss. Right of zero: the forecast was better. One dot per market.

**How to read it.** Nearly every dot is right of zero, as the research predicts: volatility clusters, and a model that weights recent days more heavily tracks it better. It still failed my test, because the gain in the first half of the history was too small to rule out luck. A forecast is useful for sizing risk, not for timing returns.

**What I did not test, and why.** Downside beta is a premium among stocks, not a way to read an index, and a later replication found it does not hold up. Hurst-exponent estimates of persistence are biased by short-run patterns (Lo, 1991), so I used variance ratios. Crash-weighted momentum needs short positions. New signal types for the trend score were left out because the common trend filters turn out to be near-equivalent (Levine and Pedersen, 2016). The [specification](https://tipsheet.markets/research/trend_models_v2_spec/) lists every reference, checked against its source.

## The third review: momentum's risk, and trend by speed

In October 2026 I tested four more ideas about stock momentum and asked whether trend signals of different speeds behave differently, prompted by a fund manager's argument that faster trend models protect better in a crisis. I wrote down every test and its pass mark first. 2 of 9 passed: momentum built from many small gains outlasted momentum built from a few jumps (narrowly), and scaling the momentum factor by its own volatility halved its worst month. Across speeds, every version of the trend score cut the worst falls and crisis losses by about two-thirds against holding; the fast signals cut them a little more, but not by enough to rule out luck, and after tax none kept up with buying and holding.

Stock momentum on the liquid point-in-time universe, 2003 to 2026. Trend speeds on 79 investable markets from each index's launch, after costs; after tax at the 30% slab. Each test was registered before its result was seen.

*Stock momentum on the liquid point-in-time universe, 2003 to 2026. Trend speeds on 79 investable markets from each index's launch, after costs; after tax at the 30% slab. Each test was registered before its result was seen.*

| Question | Result | The deciding number |
| --- | --- | --- |
| Does momentum built from many small gains outlast momentum built from a few jumps? | Passed | +0.88% a month (p 0.0121; bar 0.0125). Positive in both halves, but 1.33% then 0.46% |
| Does that measure, added as a seventh flavour, improve the blend? | Failed | −0.02% a month against the six-flavour blend (p 0.64); it stays out |
| Does scaling the momentum factor by its own recent volatility help? | Passed | Sharpe 1.02 to 1.43 (95% interval of the gain +0.15 to +0.60); worst month −45% to −24% at the same volatility |
| Is momentum from 12 to 7 months ago stronger than from 6 to 2? | Failed | No: 0.94% against 1.40% a month. Recent momentum is the stronger one here, as studies outside the US have found |
| Do fast trend signals (one and two months) cut the worst fall more than slow ones (nine and twelve)? | Failed | Shallower by 4.2 points at the median, on 75% of 79 markets; p 0.026, short of the corrected bar |
| In the Nifty 500's worst months, do fast signals lose less than slow ones? | Failed | −1.9% against −2.5% a month (holding: −6.5%); fast minus slow +0.57 points, 95% interval −0.32 to +1.54 |
| Does tax take more from the fast signals? | Failed | Before tax fast led slow by 0.9 points a year; after tax it trailed by 0.2, on 52% of markets (two-thirds needed) |
| Does tilting the score toward fast signals (50/30/20) beat the equal blend? | Failed | Sharpe +0.011 at the median (95% interval −0.033 to +0.058) |
| Does any speed beat buying and holding? | Failed | Best deflated Sharpe 0.38 with 111 trials; after tax every speed trailed holding by about 3 points a year |

### In the worst months every speed lost about a third of what holding lost

Average monthly return of each long-or-cash rule, equal weight across 79 markets, in each fifth of the Nifty 500's months (65 months each), 1999 to 2026, after costs.

**How to read it.** Left, the market's worst months; right, its best. A rule that sits in cash loses less on the left and gains less on the right. The fast signals lose least in the worst fifth and gain most in the best, but the gaps between speeds are small beside the gap to holding.

Fast All 24 signals Slow Buying and holding

In the worst months every speed lost about a third of what holding lost. Dot plot of 5 items. Fast (1–2 months): highest Best fifth (Nifty 500 10.0%) at 4.9%, lowest Worst fifth (Nifty 500 −8.2%) at −1.9%.

**Corrected 5 Nov 2026.** The after-tax replay traded a session later than the pretax backtest, so these trend-speed results were re-run with the trades on the same day, over the same window (to 13 Oct 2026). The original input snapshot no longer exists, so the re-run reads the inputs pinned on 29 October 2026, which were revised after 15 October and move the pretax figures slightly too. On the Nifty 500, the fast rule's after-tax return was 11.5% a year in the original frozen results, 11.5% on the revised inputs with the old timing, and 12.5% corrected. Every result in the table is the same as first graded: fast still beat slow before tax and trailed it after tax in 53% of markets first, 52% now, short of the two-thirds the test needed.

**What changed on the site.** The trend pages now show each market's fast, medium and slow thirds of the score, and flag when the fast and slow signals point opposite ways; that is a description, not a signal. The momentum section carries the factor's volatility reading and the continuous-against-discrete reading. The seventh flavour stays out of the blend. The [specification](https://tipsheet.markets/research/momentum_v3_spec/) has every definition, reference and number.

## The workings

### Method

- [The single rules, costs and tax](https://tipsheet.markets/methods/models/)
- [The ensemble and momentum tests](https://tipsheet.markets/methods/barometers/)
- [Trend, momentum and regimes](https://tipsheet.markets/methods/trend_momentum/)

### Data

Every chart has a table view, a CSV download and a link to its published JSON. Bundles are listed in [data status](https://tipsheet.markets/status/).

### Read with

- [Trend and momentum of every market today](https://tipsheet.markets/trend/)
- [Pre-registered: trend rules and sector momentum](https://tipsheet.markets/research/trend_momentum_spec/)
- [Pre-registered: the trend ensemble](https://tipsheet.markets/research/trend_barometer_spec/)
- [Pre-registered: stock momentum](https://tipsheet.markets/research/momentum_barometer_spec/)
- [Pre-registered: the third review, momentum and trend by speed](https://tipsheet.markets/research/momentum_v3_spec/)
- [Pre-registered: the second review of trend, momentum and risk models](https://tipsheet.markets/research/trend_models_v2_spec/)
- [Asset mixes after tax](https://tipsheet.markets/lab/)

---

Historical research, not investment advice. This page in HTML: https://tipsheet.markets/trend/evidence/. Every published dataset: https://tipsheet.markets/data/v1/manifest.json. A map of the site: https://tipsheet.markets/llms.txt
