Why do anomalies disappear after they get detected?

Question

In financial markets, anomalies refer to situations when a security or group of securities performs contrary to the notion of efficient markets, where security prices are said to reflect all available information at any point in time.

There is some evidence however that specific anomalies may decay in repeated experimental markets. Two studies showing this result include: Shogren et al (2001) who report erosion of the gap between willingness to pay (WTP) and willingness to accept (WTA) when values of goods are elicited in repeated auctions; and Cox and Grether (1996) who report decay of preference reversal when lottery values are elicited in repeated auctions.

What are other explanations of why an anomaly would disappear? How would this technically work?

Answer 1

The best explanation I have seen so far is the so-called Adaptive Market Hypothesis by Andrew Lo:

The adaptive market hypothesis, as proposed by Andrew Lo, is an attempt to reconcile economic theories based on the efficient market hypothesis (which implies that markets are efficient) with behavioral economics, by applying the principles of evolution to financial interactions: competition, adaptation and natural selection.

It has several implications, esp. 2. and 5. answer your question:

1. To the extent that a relation between risk and reward exists, it is unlikely to be stable over time.
2. There are opportunities for arbitrage.
3. Investment strategies—including quantitatively, fundamentally and technically based methods—will perform well in certain environments and poorly in others.
4. The primary objective is survival; profit and utility maximization are secondary.
5. The key to survival is innovation: as the risk/reward relation varies, the better way of achieving a consistent level of expected returns is to adapt to changing market conditions.

EDIT: Another, unfortunately very realistic, possibility is that many anomalies didn't exist in the first place but that they were just spurious (a.k.a. "fooled by randomness")...

Answer 2

Any anomaly that can be phrased as a "mispricing" or "relative value" opportunity can be expected to disappear as more people discover it and trade on it.

For example, say that stock movements over the last 15 minutes of the day are found to be strongly mean-reverting. That is, stocks which decline over the last 15 minutes of the day tend to be undervalued, and stocks which advance over the last 15 minutes tend to be overvalued.

A trader who wants to take advantage of this needs to buy stocks that have declined, and sell stocks that have advanced. That buying and selling activity pushes up the price of the stocks that declined (making them less undervalued) and pushes down the price of stocks that advanced (making them less overvalued) and thereby reducing the size of the opportunity for other traders.

In some cases, you may not have an "anomaly" but instead a "risk premium". The difference is that an anomaly has positive expected return because other market participants have not yet caught on to it, whereas a risk premium has a positive expected return because it is the reward demanded by the market for bearing a risk. For example, the "value risk premium" is generally assumed to be a compensation for risk, since value stocks often look riskier (high leverage, high earnings volatility, may be distressed).

Answer 3

Adding to the excellent answer of vonjd:

another, more cynical interpretation is that some of these "anomalies" never would have existed ex-post and that their discovery is the inevitable result of thousands of researchers looking for patterns in the same data set.

Look at ". . . and the Cross-Section of Expected Returns" by Harvey et. al. for an elaboration on this idea and "The Probability of Backtest Overfitting" by Bailey et al.

Reading both will give you some healthy skepticism regarding anomalies.

Answer 4

In high-frequency energy trading, it's possible for a trader to have hundreds of orders in the book, possibly on more than one exchange, e.g. ICE and NYMEX.

A trader may be running dozens of algorithms simultaneously. Each of these may respond independently to the changes it detects in the market data, as well as to the timing and fill data from its own execution reports.

In some situations, a sudden price change in an influential futures market may require the modification of order prices on both sides of the market over dozens of delivery periods. And then of course there are spreads, butterflies and other futures strategies that link the prices of their legs. Orders in these instruments may also have to be adjusted.

It's important to realize that even within an exchange, instruments related in the economic sense are not necessarily traded on the same match engine, and that execution reports (to individual traders) travel along different paths than market data (published generically to all subscribers). A multi-product strategy must take this into account. NYMEX and ICE also trade financially-settled equivalents to each other's physically settled contracts, and these take time to change as well.

Algorithms vary in terms of how much data they require in order to detect and act on a price change. The algorithm operators can also run them differently on days when the market is expected to be volatile, e.g. with thresholds that reduce the risk of a misplaced execution.

Even a trader with black boxes co-located at the exchanges still needs time to transmit the necessary cancel/replace messages, especially since other traders will be attempting to do the same thing. A mispricing in a set of linked markets can therefore exist for several milliseconds, and in some cases as much as several hundred milliseconds, i.e. the time needed for the majority of algorithmic participants to update their local book states and begin waiting for the next triggering event.