Momentum, Market Rotation and Holding Winners: A Review of Our Research

The questions behind the research

The starting observation was that market leadership often persists for months. Once a theme attracts sustained investment, buying after the initial advance can still produce substantial returns. Finding the exact bottom should not be necessary.

We wanted to establish whether simple rules could capture this behaviour. Could industry strength help identify stocks with better prospects? Would volume or a more selective entry improve returns? How much room should an existing holding receive before replacement?

Another observation concerned corrections. A strong market can fall, experience a volatility spike and then resume its advance. We tested whether waiting for volatility to decline improved entry after a sell-off.

We also considered changing market conditions. Rules that worked during one period might perform differently under tighter liquidity, higher inflation or different forms of institutional participation. Systematic funds and options hedging could contribute to trends and reversals, but most of our available data measured prices rather than the orders behind them.

These observations became research questions. They were not treated as established explanations.

Reading the results

The tables contain three different measures.

Portfolio returns include allocation, trading and compounding. Annualised return is the compound growth rate over the stated period.

Event returns measure performance after individual signals. Events may overlap and do not necessarily represent an executable portfolio.

Return differences compare groups, such as stocks in strong industries against similar stocks elsewhere. A difference of one percentage point is not a one per cent annual portfolio return. Tables abbreviate percentage points as pp.

Maximum drawdown measures the largest decline from a previous portfolio peak. Within-episode drawdown measures losses during an individual test window.

Test periods differ. Most portfolio comparisons assumed costs of 0.10% per dollar bought or sold. Selected tests also used 0.25%. These assumptions do not reproduce every investor’s commissions, spreads, taxes or currency costs.

The latest observations across the research extend to 13 October 2026. Stock tests using historical S&P 500 membership end on 18 August 2026.

Industry labels and forward returns

The original industry framework combined relative strength, breadth improvement and return acceleration. Labels included early turns, confirming leadership, momentum extensions, late rollovers and persistent laggards.

The original study covered 109 industries from 2019 to July 2026.

These were gross forward spreads, not net portfolio returns. Both uncertainty intervals included zero.

Stage rankings were unstable across periods. Persistent laggards did not consistently continue underperforming, and early turns did not reliably become sustained leaders.

We tested the first early-turn signal separately. Across 731 matched observations, the median 40-session advantage over comparable momentum observations was approximately −0.34 percentage points.

Waiting for confirming leadership did not improve the primary comparison. Only about 8% of qualifying early-turn episodes reached that label within 20 sessions. The waiting strategy consequently spent much of the period in cash.

The labels described changes in market conditions. The tests did not establish a reliable allocation rule based on them alone.

Market conditions and regime dependence

A concern throughout the research was that averaging across different environments could hide a useful conditional result.

The original study included weeks resembling a reference market environment. Selection used a positive broad-market trend and measures of returns, volatility, breadth and dispersion.

The subset looked stronger, but it was retrospective. It was not an independent evaluation of a regime rule specified before examining the outcomes.

This distinction addresses the idea that markets changed after major crises and may continue changing with institutional participation and technology. It is reasonable to investigate different environments. It is also easy to select historical conditions that make a rule look better.

We did not establish a validated switching rule based on inflation, liquidity, credit spreads or policy conditions.

Strong stocks within strong industries

We next tested whether industry strength improved the outcome of an existing stock setup.

Strong context required positive industry relative performance and broad participation. The stock under evaluation was excluded from its own peer calculation.

The initial result was encouraging. The longer test did not clearly confirm it.

The longer study used a price-based approximation and different coverage, so it was not an exact replication. It nevertheless reduced confidence that the initial result represented a dependable advantage.

A separate downside test did not establish that strong industry context protected against a stock falling more than 10% below entry.

Industry strength remains relevant context. These tests did not validate it as a return or downside filter.

Volume and quiet consolidation

We examined two distinct volume hypotheses.

The first required volume expansion alongside an existing momentum setup. The second required relatively quiet trading before a breakout.

The blanket volume requirement did not improve returns.

Quiet consolidation produced positive estimates, but both primary uncertainty intervals included zero. There was no clear improvement in adverse price movement or early breakout failure.

These were signal studies using a later-selected stock universe. They excluded execution costs and did not simulate a complete portfolio.

Neither test established sufficient evidence for a trading rule.

Sector momentum over a longer history

To include earlier market conditions, we tested four simple rules on the nine original US sector ETFs from January 2001 to October 2026.

Portfolios selected three sectors monthly. The six-month ranking used the preceding 126 sessions. The twelve-month ranking excluded the latest month.

Trend versions required price above the 200-day average and used weekly closing signals for exits. Transactions occurred at the following session’s close.

Twelve-month momentum modestly outperformed in the primary comparison. At costs of 0.25% per traded dollar, its annualised return fell to 9.08%, compared with 9.13% for SPY.

Trend exits reduced drawdowns but substantially reduced returns. Adding a hypothetical return on cash did not reverse that conclusion.

The longer history therefore produced a small, cost-sensitive result for one specification, rather than consistent outperformance from sector rotation.

Cross-market allocation

We tested momentum across US equities, developed and emerging international equities, gold, Treasuries and commodities.

The strategy selected two assets using six-month returns. A second version held cash when a selected asset was below its ten-month moving average.

Period: January 2008 to October 2026.

The main benefit was lower historical drawdown. Neither version exceeded SPY’s compound return over the full period.

The asset universe mattered. Removing gold reduced returns, while removing Treasuries substantially increased drawdown in the unfiltered strategy.

This supported a possible risk-management application, rather than a demonstrated return advantage.

Bitcoin timing rules

The Bitcoin tests asked whether simple trend rules could retain much of the upside while reducing prolonged losses.

Signals used completed daily or weekly observations and executed at the following daily close. Simulations used spot Bitcoin, no leverage, zero interest on cash and trading costs.

Period: January 2019 to October 2026.

Some comparisons changed with the starting date. A small return advantage for the 200-day rule over a longer window disappeared under higher costs.

Several rules reduced drawdowns while sacrificing returns. None consistently dominated across the comparisons. These historical annualised returns are not forecasts.

We did not establish a valid timing rule based on a fixed number of days after a final market top or bottom. Such a rule requires a definition of the turning point that can be applied before the subsequent price path is known.

Entry after a volatility spike

The hypothesis was that buying after selling pressure eased might improve the outcome of entering during a correction.

We compared immediate entry after a qualifying SPY sell-off with waiting for declining realised volatility or a recovery above the 20-day moving average.

The study contained 18 non-overlapping episodes. Each approach was measured to the same six-month endpoint.

The volatility rule’s average return advantage was approximately 0.29 percentage points. Excluding the COVID episode changed it to a disadvantage of approximately 0.99 percentage points.

It did not improve the worst drawdown in the sampled SPY episodes.

The episode design omitted some later crisis legs. These figures are not a complete portfolio simulation or a full measure of crash protection.

Positioning and systematic flows

We examined a public positioning proxy to test part of the idea that easing institutional selling might improve recovery entries.

The study combined SPY recovery signals with CFTC E-mini futures positioning.

The selected groups were too small for a credible conclusion.

CFTC positioning is not a direct record of CTA orders or dealer hedging. The test also used revised historical reports and conservative publication-lag assumptions rather than exact original release reconstruction.

The wider ideas about systematic buying, options-related volatility suppression, Treasury issuance, collateral demand and liquidity were not verified as causal explanations by these tests.

That limits the conclusion. A price-based result cannot establish which institution or trading mechanism caused it.

Holding rules and profit-taking

A separate systematic stock study compared holding for one year with a 200-day trend exit and re-entry policy. A third version added a partial sale after the stock doubled.

The study contained 2,343 paired stock episodes.

Trend exits reduced large individual losses but also reduced exposure and returns. The partial-sale rule further reduced average returns.

These were overlapping stock episodes, not portfolio returns or portfolio-loss probabilities. The universe also contained survivorship limitations.

The test measured the cost and benefit of specific exit rules. It did not establish that every stock should be held indefinitely.

Concentrated stock leadership

The stock portfolio research addressed the observation that a small group of companies can generate exceptional returns during a sustained theme.

Initial tests used a recently assembled stock list. Their strong results were not accepted as evidence of an edge because the candidate universe benefited from hindsight.

We repeated the test using historical S&P 500 membership and available Yahoo prices.

The baseline selected the ten eligible stocks with the strongest positive six-month returns each month. Positions were equally weighted, with execution at the next session’s close.

One variant required a compact consolidation breakout before entry. Another retained existing holdings while they remained within the top twenty and maintained positive momentum.

Period: January 2019 to 18 August 2026.

The entry filter reduced participation. Its lower drawdown did not establish better entry selection.

The retention rule improved both return and drawdown in this sample without reducing average equity exposure. It allowed an existing holding to lose some relative rank without immediately being sold.

Monthly review versus monthly replacement

Across 90 monthly comparisons, newly selected stocks underperformed removed stocks by approximately 2.2 percentage points over the following 21 sessions, on average.

This was a diagnostic comparison, not a separate compounded portfolio.

It suggested that review frequency and replacement criteria should be considered separately. A monthly review need not require selling every holding that falls slightly in the ranking.

Under the original rule, a stock slipping from eighth to twelfth would be sold. Under the retention rule, it could remain until it fell outside the top twenty, lost positive momentum or otherwise failed eligibility.

The number of entries fell from 447 to 307. The median completed holding period increased from 22 to 44 trading sessions.

Performance was uneven.

The 2026 result was consistent with the observation that sustained thematic leadership can produce large gains without buying the exact bottom.

However, Micron, Seagate, SanDisk and Western Digital contributed approximately 79 percentage points to the retention portfolio’s 2026 return before trading fees. This represented concentrated exposure to a related memory and storage cycle.

At costs of 0.25% per traded dollar, the retention portfolio’s full-period annualised return fell to 22.90%, compared with 22.93% for QQQ. The average monthly improvement over the original strategy also remained statistically uncertain.

The rule deserves further evaluation, but the current evidence does not establish dependable outperformance.

Research that did not reach a performance test

Earnings-reaction continuation reached a data-feasibility assessment. Reliable event timing, historical eligibility and missing return histories remained unresolved. It was not a completed broad strategy test and cannot be counted as either a demonstrated success or failure.

Whale accumulation and distribution were also not established as tested strategies. Trade size does not identify an institution, and aggregate price or volume data do not reliably establish the initiating side of individual trades.

Discussion of dealer positioning, CTA activity and liquidity should therefore be distinguished from direct measurement of those mechanisms.

What the research supports

Industry labels described market conditions but did not establish a reliable allocation advantage. Strong industry context produced encouraging short-sample results that were not clearly confirmed over a longer period.

Trend rules reduced some drawdowns, generally at the cost of participation and returns. Volume confirmation and restrictive entries did not consistently improve outcomes. Volatility-based recovery timing depended materially on one exceptional episode.

Retaining established stock leaders produced the strongest recent portfolio result. It also relied heavily on one year, related stocks and assumed trading costs.

The concern about changing market regimes remains unresolved. Splitting historical data into increasingly specific environments can reveal differences, but it also creates more opportunities to fit rules to past outcomes.

Data limitations remain material. Historical membership improved the stock tests, but some former constituents lacked usable prices. Exceptional returns were examined through sensitivity checks rather than automatically deleted, since large genuine gains and losses are part of the strategy’s economics.

The studies also reused overlapping history. Their findings are exploratory, and the reported uncertainty does not fully account for the number of ideas examined.

The most specific result concerns ownership after selection. A stock can stop meeting the standard for a new purchase while remaining eligible to hold. Our retention test supports examining that distinction. It does not yet establish a reliable advantage over buying and holding.