Trend Following vs Counter-Trend and How Statistical Pairs Trading Works

Most trading approaches belong to one of two families. One assumes that a move that has started will continue, and the other assumes that a stretched move will snap back. The two have different shapes of results, different ways of failing and different demands on the person using them. A third approach, statistical pairs trading, applies the second idea to the gap between two instruments instead of to a single price.

Trend following

A trend follower buys when price is rising and sells when it is falling, and does not try to call the turn. Entries often come from a breakout to a new high, a moving average crossover or a pullback in an established trend. The stop is usually trailed behind price, so that the trade is closed only when the trend gives back a set amount. See trends, trendlines and channels for the chart side.

The typical profile is a low win rate with winners that are several times larger than the losers. Many trades are stopped out in sideways markets for small losses, and a few ride a long move for a large gain. The hard part is behavioural: sitting through a run of small losses, and giving back part of an open profit before the trailing stop is hit.

Counter-trend trading

A counter-trend or mean-reversion trader sells when price has risen unusually far and buys when it has fallen unusually far, expecting a return toward an average. Common tools are oscillators such as the RSI, bands such as Bollinger Bands, and the edges of a range or channel.

The profile is the mirror image: a high win rate, small gains, and occasional losses that are larger than the gains. It feels comfortable because most trades win. It fails when a stretched market keeps stretching, and a strong trend can turn many small wins into one loss that wipes them out. A counter-trend trade without a hard stop is the most common way this style blows up an account.

Side by side

Win rate is not the measure

Neither profile is better by its win rate. What counts is expectancy, the average result per trade in units of the amount risked. The numbers below are invented for illustration.

In this example the system that loses nearly two trades in three is the profitable one, and the system that wins seven in ten loses money. A real counter-trend system can have a positive expectancy too. The point is that you cannot judge it from the win rate alone, and the risk per trade and expectancy article shows how to calculate it from your own records.

Telling the market regime apart

Markets alternate between trending and ranging, and there is no reliable way to know which phase is next. Some traders use a trend-strength measure such as the ADX, with a reading above 25 commonly taken as a trending market and below 20 as a ranging one. Those thresholds are conventions, and the indicator lags. The multi-indicator system article covers how it is used.

A practical way to cope is to choose the style that suits your temperament, test it over different market phases, and accept that it will have bad stretches. Switching styles after each losing run tends to put you in the wrong one at the wrong time.

What statistical pairs trading is

Pairs trading takes two instruments whose prices have historically moved together, such as two companies in the same sector. When the gap between them becomes unusually wide, the trader sells the one that has risen relative to the other and buys the one that has lagged. The bet is on the gap closing, not on the direction of either price, and for that reason it is often described as market neutral. It is closer to the counter-trend family, applied to a spread.

Academic work gives the idea a basis. Gatev, Goetzmann and Rouwenhorst tested a simple pairs rule on US stocks from 1962 to 2002 and reported positive excess returns after allowing for costs. Later work by Do and Faff (Financial Analysts Journal, 2010) found that the profits of simple pairs trading kept shrinking in US data to 2009, though the strategy held up better in turbulent markets. A rule that is widely known and easy to copy tends to lose its edge, so treat the original result as evidence that the idea can work, not that it will.

Cointegration, the spread and the z-score

Two price series are cointegrated when a combination of them, typically the difference or ratio, stays within a stable range even though each series wanders on its own. The concept comes from the econometricians Robert Engle and Clive Granger, whose work on it was recognised with a Nobel Prize in 2003. Two stocks can both trend upward and not be cointegrated, and two stocks can look quite unlike each other and be. Traders usually test for it with a statistical test such as the augmented Dickey-Fuller test on the spread.

The spread is that combination. Its z-score says how far it is from normal, measured in standard deviations over a lookback period.

A z-score of 0 means the spread is at its average. A reading of +2 means it is two standard deviations above, which, if the spread were normally distributed, would be unusual. Many traders enter at +2 or -2, exit near 0 and cut the trade at +3 or -3 or after a set time. Those levels are conventions, and real spreads have fatter tails than a normal distribution.

A worked example

Take two invented stocks in the same sector, A and B, with a lookback average ratio of 2.00 and a standard deviation of 0.10. A trades at 44.00 and B at 20.00, so the ratio is 44 / 20 = 2.20.

The two legs have equal money value, so a move in the whole sector affects them in similar ways. Suppose the ratio reverts, with A at 41.00 and B at 20.50, so that 41.00 / 20.50 = 2.00.

Now suppose it goes the other way, with a stop at a z-score of +3. That is a ratio of 2.30. With B unchanged at 20.00, A would be at 46.00, and the short in A loses 250 x 2.00 = 500. If you did not stop out and the relationship broke, with A at 52.00 and B still at 20.00 (a ratio of 2.60 and a z-score of +6), the loss would be 250 x 8.00 = 2,000, nearly twice the profit in the good case. That asymmetry is the central risk.

What can go wrong

• The relationship breaks. A takeover, a profit warning, a regulatory change or a change in the business can end a historical link permanently. The spread then does not come back.
• The statistics are unstable. The average and standard deviation depend on the lookback you choose, and a pair that passed a test in the past may not pass it next year.
• Overfitting. Searching hundreds of pairs for ones that fit the past will find some by chance. See the next article on testing claims.
• Costs. There are two spreads or commissions on entry and two on exit, plus borrowing costs on the short leg.
• Not truly neutral. Equal money value does not mean equal sensitivity to the market, so a market-wide move can still help or hurt.
• Leverage. Two positions on margin double the exposure and the financing, and losses on a short are not capped.

Shorting a stock or an index is possible through CFDs and short selling, each with its own costs and rules. Pairs trading suits people who are comfortable with statistics and can check the assumptions again each month. It is not a lower-risk shortcut, and the stop at a fixed z-score or a fixed loss is as necessary here as anywhere. See where to place stop losses.

Choosing between them

The three approaches ask for different things. Trend following asks for patience through losing runs. Counter-trend trading asks for discipline to take a loss when a trade that usually works does not. Pairs trading asks for statistical care and a plan for the day a relationship ends. Pick the one whose failure you could live with, and test it with small size before committing real money.

Sources: Gatev, Goetzmann and Rouwenhorst, Pairs Trading: Performance of a Relative-Value Arbitrage Rule, Review of Financial Studies, 2006; Do and Faff, Does Simple Pairs Trading Still Work?, Financial Analysts Journal, 2010; Engle and Granger on cointegration (Nobel Prize in Economic Sciences, 2003). The expectancy and pairs examples are invented. Next in the series: Algorithmic and AI-Assisted Trading: Systematic Rules.