Trend Following and Momentum Research: The Anomaly That Refuses to Die

Of all the patterns that challenge the clean theory of efficient markets, momentum is probably the most stubborn. It's been documented across centuries of data, dozens of countries, and nearly every asset class anyone has bothered to check — stocks, bonds, currencies, commodities, even cryptocurrencies. And yet, by the logic laid out in the first post of this series, it shouldn't really exist at all: if past price behavior predicted future price behavior, traders would exploit that pattern until it vanished. It hasn't.

This post digs into what momentum and trend following actually are, the research behind them, the competing explanations for why they work, and the very real risks that have humbled trend-followers more than once.

Momentum vs. Trend Following: Two Related but Distinct Ideas

These terms get used almost interchangeably in casual conversation, but they describe slightly different things, and it's worth being precise.

Momentum is a relative concept, typically studied in academic finance as a cross-sectional phenomenon: rank a universe of assets by their recent performance, and the assets that performed best over some recent lookback period (commonly 3 to 12 months) tend to continue outperforming the assets that performed worst, over the following weeks or months. Momentum strategies are usually long the winners and short the losers — a relative bet, not a directional one on the market as a whole.

Trend following is more often an absolute (time-series) concept: it asks whether a single asset's own recent price trend tends to persist, independent of how it compares to anything else. A trend-following strategy typically goes long an asset that's been rising and short (or flat) one that's been falling, based purely on its own price history, often using tools like moving averages or breakout rules to define "trending."

In practice, these two ideas overlap heavily and the underlying research often supports both — the line between them is more a matter of how a strategy is constructed (relative ranking versus absolute trend) than a sharp theoretical divide.

The Foundational Research

Jegadeesh and Titman (1993)

The single most influential academic paper here is Narasimhan Jegadeesh and Sheridan Titman's 1993 study, which systematically tested strategies that bought stocks with strong returns over the prior 3-12 months and sold stocks with weak returns over the same period, then held those positions for a further 3-12 months. They found these strategies generated statistically and economically significant excess returns in U.S. equities — a result that landed awkwardly for semi-strong market efficiency, since it implied a systematic, exploitable pattern derived purely from public price history (arguably closer to a weak-form violation, since it uses only past prices).

This finding triggered a wave of follow-up research asking the obvious question: is this a fluke, a statistical artifact, or something real?

Out-of-Sample and Out-of-Market Confirmation

What made momentum hard to dismiss as data mining was how it held up under scrutiny that usually kills market anomalies:

It was tested out of sample, on data after the original study period, and the effect persisted.
It was tested in international equity markets outside the U.S., and the effect generally showed up there too.
It was tested in completely different asset classes — commodities, currencies, government bonds — by researchers including Geert Rouwenhorst and, later, in a particularly influential and comprehensive 2013 study by Tobias Moskowitz, Yao Hua Ooi, and Lasse Heje Pedersen titled "Time Series Momentum," which examined trend-following across 58 different liquid futures markets spanning equity indices, bonds, currencies, and commodities going back decades.
It was even tested on very long historical datasets reconstructed back over a century, in some cases, and the pattern still appeared.

This breadth is exactly why momentum is treated so seriously in academic finance, despite being uncomfortable for clean efficiency theory: a pattern confined to one narrow dataset is easy to dismiss as overfitting or luck; a pattern that shows up independently across asset classes, countries, and centuries is much harder to wave away.

Why Would Momentum Exist? Competing Explanations

There isn't full consensus on why momentum works, and the explanations on offer say a lot about the broader efficient-markets-versus-behavioral-finance debate from earlier posts in this series.

Behavioral Explanations

The most common behavioral story involves a two-stage process:

Underreaction. When genuinely new information arrives — an earnings beat, a positive product announcement — investors don't fully and immediately price it in. This can happen because of anchoring (investors are slow to move far from a prior belief), conservatism bias (a documented tendency to underweight new evidence relative to existing views), or simply because the information diffuses gradually through the market rather than reaching everyone simultaneously. The post-earnings-announcement drift discussed in the first post of this series is closely related to this same underreaction mechanism.
Subsequent overreaction. As the price drifts upward on continued buying, the trend itself becomes a piece of information — late arrivals see the rising price and herd in, sometimes pushing the price beyond what fundamentals justify, which sets up the later reversal (mean reversion) that's also documented in longer-horizon studies.

Momentum, in this account, is essentially a slow-motion overreaction-correction cycle, driven by the same biases (herding, anchoring, underreaction to new information) discussed in the behavioral finance post.

Risk-Based Explanations

Not everyone accepts the behavioral story. A more efficiency-friendly explanation argues that momentum returns are actually compensation for bearing some real, but hard-to-measure, risk — meaning momentum isn't "free money," it's a fair reward for a genuine economic exposure that simply hasn't been fully identified or modeled yet. Under this view, momentum strategies occasionally suffer brutal, sudden losses (more on this below) precisely because they're exposed to a real risk factor, not despite it — the strategy's long-run excess return is the market's price for occasionally bearing that risk.

Limits to Arbitrage

A third, complementary explanation echoes the earlier microstructure and behavioral posts: even if momentum is partly behavioral and partly exploitable, the limits-to-arbitrage problem may prevent it from being fully traded away. Momentum strategies, particularly when many funds run similar versions simultaneously, can become crowded and exposed to violent, simultaneous unwinds (discussed below) — a real cost of trying to exploit the anomaly that helps explain why it hasn't disappeared even though it's been public knowledge in academic literature for over three decades.

Most serious researchers today would say the honest answer is "some mix of all three," with the exact blend still debated.

How Trend-Following Strategies Are Actually Built

Trend following has a long practical history that predates the academic research significantly — it's one of the oldest systematic trading approaches, used by Commodity Trading Advisors (CTAs) and managed futures funds for decades, well before Jegadeesh and Titman's paper gave it academic legitimacy. The mechanics typically involve:

Moving average crossovers — comparing a shorter-term moving average (say, 50-day) to a longer-term one (say, 200-day); a "golden cross" (short above long) signals an uptrend, a "death cross" signals a downtrend.
Breakout systems — entering a position when price exceeds its recent trading range, on the logic that a genuine breakout tends to continue (this is the core logic behind the classic Donchian Channel and "Turtle Trading" systems from the 1980s).
Time-series momentum signals — simpler still: is the asset's return over the past N months positive or negative? Go long if positive, short (or flat) if negative.
Risk-adjusted / volatility-scaled positioning — rather than betting equal dollar amounts on every signal, many systematic trend strategies size positions inversely to recent volatility, so a calm, steadily trending bond market and a wildly swinging commodity contribute similar risk to the overall portfolio rather than the most volatile asset dominating the result.

A genuinely distinctive feature of trend-following, compared to most other investment styles, is that it's agnostic about direction — it doesn't have a structural long bias. A well-built trend system is just as willing to go short a falling market as long a rising one, which is part of why these strategies have historically tended to perform well during sustained bear markets and crises (1973-74, 2000-02, 2008) when most directional, long-only strategies suffered badly.

Diversification Properties: Why Institutions Care

One major reason trend following gets serious institutional attention, beyond the raw historical returns, is its correlation profile. Trend-following strategies have, across long historical samples, tended to show low or even negative correlation to traditional stock and bond portfolios, particularly during major equity bear markets — precisely the environment where most other diversifiers (including many hedge fund strategies) tend to fail when investors need them most.

This "crisis alpha" property — performing relatively well exactly when traditional portfolios are struggling — is the primary reason large pension funds, endowments, and multi-strategy investors allocate to managed futures and trend-following programs, even when the standalone returns in calm, steadily rising markets can look unimpressive or even negative for extended stretches.

The Brutal Risk: Momentum Crashes

Momentum's biggest, most well-documented weakness deserves its own discussion, because it's exactly the kind of risk that gets hidden in long-run average return statistics.

A landmark paper by Kent Daniel and Tobias Moskowitz, fittingly titled "Momentum Crashes," documented that momentum strategies, despite generally strong long-run average returns, are subject to sudden, severe, and clustered crashes — most dramatically during sharp market rebounds following a steep downturn. The mechanism is intuitive once you see it: a momentum strategy that's been short the prior losers (often beaten-down, high-risk, high-volatility stocks) gets badly hurt if those same beaten-down stocks suddenly and violently rebound — exactly the pattern seen, for instance, in the sharp market recovery following the 2009 market bottom, and again during similar rapid reversals.

This matters enormously for how the strategy should actually be understood: momentum's return distribution is negatively skewed — meaning it tends to deliver many small, steady gains punctuated by occasional severe, sharp losses — rather than behaving like a smooth, normally distributed return stream. An investor who only studies the long-run average return, without understanding this skew, will badly underestimate the strategy's real risk and will likely abandon it at precisely the worst moment, right after one of these crash episodes, rather than holding through it as the long-run statistics would technically justify.

Why Hasn't Arbitrage Erased Momentum?

This is the most theoretically interesting question in the entire topic, and it ties directly back to the limits-to-arbitrage discussion from the market efficiency post. A genuinely persistent, well-documented anomaly known for over 30 years should, in principle, attract enough capital to trade it away. Several explanations have been proposed for why that hasn't fully happened:

The crash risk itself deters capital. Because momentum crashes are sudden, severe, and hard to predict in advance, the strategy is genuinely riskier than its average return alone suggests — meaning the persistent excess return may be a fair, not free, compensation for bearing that crash risk, as the risk-based explanation above suggests.
Implementation costs are real and non-trivial. Momentum strategies require frequent rebalancing (since "what was a recent winner" changes constantly), generating transaction costs, market impact (tying back to the microstructure post), and — in taxable accounts — significant short-term capital gains tax drag, all of which eat meaningfully into the strategy's theoretical, pre-cost return.
Capacity constraints. As more capital chases the same signal, momentum strategies can become "crowded trades" — many funds holding similar positions — which both compresses the available excess return and increases the risk of a simultaneous, self-reinforcing unwind if multiple large players need to exit at once (a dynamic that contributed to the August 2007 "quant quake," when several quantitative equity funds running similar momentum-related strategies suffered sudden, correlated losses over just a few days).
Behavioral persistence. If the underreaction-then-overreaction story above is correct, momentum is fed by recurring human psychology (anchoring, herding, slow information diffusion) that doesn't go away just because the resulting pattern has been published in finance journals — knowing about a bias intellectually, as the behavioral finance post discussed, doesn't automatically prevent the underlying behavior that creates it.

How Momentum Connects to the Rest of the Series

Momentum sits right at the intersection of everything discussed so far in this series:

It's a direct, empirically robust challenge to weak-form and semi-strong market efficiency, since it implies past price information (and, in the related post-earnings-drift literature, past public news) has predictive power over future returns — the opposite of what clean EMH would predict.
Its leading explanation is essentially a behavioral finance story — underreaction driven by anchoring and slow information diffusion, followed by herding-driven overreaction — making it one of the cleanest real-world bridges between the academic anomaly literature and the psychological biases discussed in the second post.
Its persistence despite being public knowledge for decades is a textbook case of limits to arbitrage, and the costs of actually trying to exploit it (rebalancing turnover, market impact, crowding, crash risk) are a direct application of the microstructure concepts from the third post — a strategy can be real and statistically robust on paper while still being genuinely difficult to capture cleanly in practice.

Practical Considerations

For investors actually considering trend-following or momentum exposure, a few grounded takeaways:

It's a real, well-documented pattern — but it's not a free lunch. Strong historical evidence across asset classes and centuries doesn't mean the strategy is easy, comfortable, or risk-free to run.
Crash risk is the central thing to understand and budget for. Momentum's negative skew means a strategy can look fantastic for years and then suffer a sudden, sharp drawdown around major market turning points — exactly when emotional pressure to abandon a strategy (tying back to the behavioral finance post's discussion of capitulation) tends to be highest.
Costs matter enormously, especially turnover and taxes. A momentum strategy's theoretical, pre-cost excess return can shrink substantially once realistic trading costs, market impact, and tax treatment of frequent short-term gains are factored in — this is exactly the kind of gap between "academic anomaly" and "investable strategy" that the microstructure post's discussion of transaction costs warned about.
Diversification value can matter more than standalone returns. For many institutional allocators, the appeal isn't "this beats the stock market" — it's the historically low correlation to stocks and bonds during crises, which is a different and arguably more durable case for an allocation than raw expected return alone.
Simplicity is not naivety. Despite the relatively simple rules underlying many trend-following systems (compare a price to a moving average; rank assets by recent return), the strategy's robustness across asset classes and long historical periods is precisely why it has survived as one of the most studied and longest-running systematic approaches in finance, even as more complex strategies have come and gone.

The Takeaway

Trend following and momentum occupy an unusual position in finance: they're simultaneously one of the most rigorously documented anomalies in the history of empirical finance research and one of the most genuinely risky, uncomfortable strategies to actually hold through a full market cycle. The pattern survives not because nobody knows about it — it's been public, published, and picked apart in academic journals for over thirty years — but because exploiting it cleanly requires bearing real risks (sudden crashes, crowding, implementation costs) that keep the opportunity from being fully arbitraged away. It's a vivid, ongoing illustration of exactly the tension this whole series has been circling: markets are efficient enough that genuine free lunches are rare, but not so efficient that every well-documented, statistically robust pattern simply disappears the moment it's discovered.

This post is for informational purposes only and isn't financial advice.

Trend Following and Momentum Research