I saw this firsthand with a volatility dispersion strategy I deployed on Bank Nifty and its constituents. What looked robust in early backtests showed a steady, unmistakable decline in returns across successive months. Not noise, not a bad period, but a directional bleed that only made sense in one frame: the edge was being competed away in real time. Alpha decay is real. It is not a theoretical concern for quants to worry about eventually. It is an active, ongoing process happening to every strategy in production right now. The moment a signal is discovered, the clock starts. Markets are learning machines, and they learn fast. That pattern in the backtest was the signal. The strategy wasn't broken. It was just becoming known.
The term borrows deliberately from nuclear physics (which is a very interesting subject in itself). A radioactive isotope loses half its activity in a fixed period — the half-life. A trading signal loses half its predictive power on a similar curve. The only question is how steep that curve is. And that answer depends almost entirely on the frequency at which you are operating.
§ 1 What Alpha Actually Decays Into
Alpha exists because of an information asymmetry — you know something the price does not yet reflect. The decay happens as that information propagates. Other participants observe the same patterns, deploy similar logic, and compete away the edge. What was once a 10 bps expected return per trade becomes 6, then 2, then noise.
This is not a failure of the strategy. It is the market doing its job. Price discovery is the mechanism that destroys edges, and it is ruthlessly efficient. The practical implication is that no signal is permanent and only signals at timescales where discovery is slow enough to still be exploitable. (I am sure EMA crossovers must have been a decent strategy in itself during 1980s.)
§ 2 HFT — Edges Measured in Microseconds
At the high-frequency end, alpha decay is not a quarterly review problem. It is a today problem. HFT strategies extract edge from execution quality, latency advantages, and order book microstructure — not from any fundamental view on where a stock should trade. The signal is the spread. The alpha is the fill.
Because the informational content of these signals is small and fleeting, they are discovered quickly and competed away aggressively. A co-location advantage is worth less every year as more participants colocate. A specific order flow pattern gets arbitraged the moment a second firm notices it. The half-life of an HFT edge can be measured in weeks to months before capacity thins out or competitors close the gap.
In HFT, the signal and the execution are inseparable. A correct directional prediction that arrives 50 microseconds late is worth zero (or less than zero). Decay here is not just about the signal fading; it is about the execution infrastructure becoming the competitive moat, and that moat shrinking as hardware costs fall and access democratises.
§ 3 MFT — Slower Decay, Deeper Roots
Medium-frequency strategies — holding periods from hours to a few days — operate on a different informational basis. The signals here are closer to price action: momentum, mean reversion, volatility regimes, order flow imbalance over longer windows. These are not pure microstructure plays. There is a fundamental logic underneath them, even if it is statistical rather than fundamental in the analyst sense.
Because the signal is rooted in something more structural, for example, how participants behave across a session, how volatility clusters, how trends persist or reverse — the decay rate is meaningfully slower. A well-constructed MFT momentum signal might have a half-life measured in months to a few years before the edge is substantially competed away. Capacity is larger, the universe of applicable instruments is wider, and the capital required to crowd it out is much greater.
| Dimension | HFT | MFT |
|---|---|---|
| Holding period | Microseconds – seconds | Hours – days |
| Edge source | Latency, microstructure, execution | Price action, flow imbalance, regimes |
| Signal nature | Mechanical, infrastructure-dependent | Statistical, somewhat fundamental |
| Alpha half-life | Weeks – months | Months – years |
| Crowding risk | Extreme — arms race dynamics | Moderate — capacity absorbs more capital |
| Decay driver | Hardware parity, competitor discovery | Capital inflows, strategy replication |
This slower decay nature of MFT is honestly one of the bigger reasons I gravitated toward it after a certain point. The other reason is more straightforward: I did not get into a CS program. HFT is an infrastructure game at its core, and without a strong CS foundation and access to the tooling that comes with it, competing on latency and execution pipelines is a non-starter. MFT let me compete on ideas and statistical rigor instead, which felt like a more honest playing field given where I was starting from.
§ 4 What To Do About It
The honest answer is: not much beyond accepting it and building for it. Walk-forward analysis, out-of-sample validation windows, and live-versus-backtest Sharpe comparisons are all decay detectors, not just performance metrics.
For HFT, the response to decay is infrastructure investment and signal refresh, not the kind of "find a new factor" research cycle that MFT shops run. For MFT, the lever is diversification across uncorrelated signals, so that as one signal decays another is still contributing. The portfolio of alphas outlives any individual alpha.
The best evidence that your alpha has decayed is often the live performance of the strategy itself (by which point you have already been trading it for months.) Building decay monitoring into production infrastructure, not just the research process, is one of the more underrated operational disciplines in systematic trading.
Alpha decay is the market's immune response to being predicted. Every strategy is in a race against it. The frequency at which you operate determines how fast you are running but also how quickly the ground beneath you shifts.