Momentum Investing

Scope. This article describes momentum as an organising belief and reports what the literature has found — in both directions. No signal, rule, ranking period, holding period or parameter is offered for use. Where the article describes how researchers constructed their tests, that is a description of research design, not a procedure this portal supplies. Momentum as a method of price analysis belongs to the technical cluster in Pillar 25; this article covers it as a school of thought. No price, multiple or valuation appears.

It is the most awkward idea in this pillar, and the awkwardness is why it matters. Every other approach here is compatible with markets working properly. Momentum is not. It says that a security's own past returns predict its future returns, which is precisely what weak-form efficiency denies. Either the evidence is wrong or one of the foundations is.

What the literature actually found

The finding entered mainstream academic work in 1993, when Jegadeesh and Titman documented that ranking US stocks on their returns over the previous three to twelve months and comparing subsequent performance produced statistically significant differences — in US common stock returns from 1965 to 1989.

The obvious objection was data mining, and it was addressed directly. The same authors returned to the question in 2001 and found the pattern had continued through the 1990s — an out-of-sample period after publication, which is the hardest test a documented anomaly faces. The pattern has since been reported in international markets, in other asset classes, over samples extended back to 1927, and in a time-series form measuring an asset against its own past rather than against its peers.

The property that the headline finding omits

Momentum's returns are not distributed the way an average suggests. Later work established that momentum portfolios occasionally suffer severe, concentrated drawdowns — characteristically when markets rebound sharply after a decline.

The mechanism is intuitive once stated. After a market fall, the securities that have fallen most become the past losers, and a strategy positioned against them is positioned against exactly the securities that rebound hardest when the market turns. The approach is therefore most exposed at the moment the market is recovering, and the losses arrive fast rather than gradually.

That changes what the average return means. A strategy with a good average and rare catastrophic episodes is a different proposition from one with the same average distributed evenly — and an investor's ability to remain in the first through an episode is not something an average measures.

The friction the approach carries by construction

Momentum requires holdings to change as relative performance changes, so turnover is not incidental to the approach — it is the approach. The arithmetic of what that costs follows directly from the holding period.

Worked example — the holding period decides the hurdle before anything else does. A one-month holding period implies twelve round trips a year, costing 4.80% of capital annually at 40 basis points — close to the whole of the canonical 5.00% equity risk premium. At a twelve-month holding period the same cost assumption produces 0.40%. The gross finding in the literature is the same in both cases; the net proposition is not remotely the same. This is arithmetic about turnover, not a performance estimate — MarketClue publishes no backtested results, and the point here is only that any discussion of momentum that omits the turnover has omitted the largest single term in the arithmetic.

Why it might happen, and why nobody is sure

The behavioural explanation holds that investors underreact to news at first and then overreact, producing continuation followed eventually by reversal. It fits the observed pattern including the long-horizon reversals, and its authors themselves have cautioned that the fit should be treated carefully.

The risk-based explanation holds that momentum returns compensate for a risk that is real but not captured by standard models — and the crash behaviour above is consistent with that, since a strategy that fails badly in recoveries is bearing something.

Neither explanation is settled, and the distinction is not academic. If momentum is compensation for risk, it should persist and it should hurt. If it is a behavioural error, it should erode as it becomes known — and thirty years after publication it is very well known.