Some of existing method of Pair trading

Pairs trading is one of Wall Streets quantitative methods of speculation which dates back to the mid-1980s (Vidyamurthy, 2004i). In its most common form, pairs trading involves forming a portfolio of two related stocks whose relative pricing is away from its equilibrium state. By going long on the relatively undervalued stock and short on the relatively overvalued stock, a profit may be made by unwinding the position upon convergence of the spread, or the measure of relative mispricingii. Pair trading is easiest to understand when considering equities can be use with any assets futures, options and currency. Especially currency are more deeply influenced by macroeconomic events than other security type and, as such, require a degree of awareness. There is many method of pairs trading. Ill mention about 2:
a) Distance trading, b) Cointegration model.

The distance method is used in Gateviii et al (1999)and Nath iv(2003) for empirical testing whereas the cointegration method is detailed in Vidyamurthy (2004). In example below I use Close price but better way is to use VWAP Volume Weighted Average Price. Distance trading

In distance trading we measure distance between two (or more) moving together assets. We can do this in different ways. Often in assets selecting process one use squared distance between normalized prices. (Equation ([1]). One could also use a maximum correlation criteria, but the results are quite similar.

) [1]

Where Pai and Pbi are normalized assets prices. Prices should be normalized because use of original prices (without normalization) would be a problem for the case of minimum squared distance rule since two assets can move together but still have a high squared distance between them. The transformation employed is the normalization of the price series based on its mean and standard deviation.

[2] Where Pit is normalized price of asset i at time t, E(Pit) is just the expectation of is the standard deviation. Pit , in this case the average, and
NZDUSD AUDUSD Close Daily Price
1

0.8

P ric e
0.6
AUDUSD NZDUSD

0.4 08-Sep-2003 00:00:00

30-Aug-2004 03:25:42

22-Aug-2005 06:51:25

14-Aug-2006 10:17:08

06-Aug-2007 13:42:51

28-Jul-2008 17:08:34

20-Jul-2009 20:34:17

13-Jul-2010 00:00:00

AUDUSD Normalized Price 5

-5 1000

1500

2000

2500

3000

3500

NZDUSD Normalized Price 5

-5 1000

1500

2000 Time

2500

3000

3500

Figure 1 Example of normalized prices

We than could use difference between this normalized prices as trade trigger . Ill write about this a little later.

difference between normalized AUDUSD NZDUSD prices 4

-1

-2

-3

-4 08-Sep-2003 00:00:00

20-Jan-2005 00:00:00

04-Jun-2006 00:00:00 Time

17-Oct-2007 00:00:00

28-Feb-2009 00:00:00

13-Jul-2010 00:00:00

Figure 2. Difference between normalized pair As we can see there is mean reversing bias we could use to trade. Some traders use normalized difference between prices another measure difference between pairs (spread between market prices) and normalize that difference (if we selected pair earlier). We can use the same equation [2] but Pit is now dit difference between prices.

[3] Where dit= Pait-Pbit

Diff between AUDUSD and NZDUSD 0.25 0.2 unnam ed 0.15 0.1 0.05 0 08-Sep-2003 00:00:00

20-Jan-2005 00:00:00

04-Jun-2006 00:00:00 Time

17-Oct-2007 00:00:00

28-Feb-2009 00:00:00

13-Jul-2010 00:00:00

normalized difference 4

2 unnam ed

-2

-4 1000

1500

2000 Time

2500

3000

3500

Figure 3. Normalized difference

One could set different test window length and mean / std.dev length ,figures above are made for mean length=20. When we know something about price and spread behavior in test period we could try to set some trading rules. We should trade (buy or sell synthetic asset buy one asset and sell another) when distance is above / below some level. How we set this levels is important. Levels could be set in different terms e.g. as standard deviation or percentile terms. In standard deviation terms we set it on some multiplier of standard deviation of test period (I mean deviation from difference between normalized prices or normalized difference of prices as describe above). In percentile terms we use distribution of our difference and calculate percentile . A percentile (or centile) is the value of a variable below which a certain percent of observations fall. So the 20th percentile is the value (or score) below which 20 percent of the observations may be found. v

Asset 1 norm 5 0 -5

200

400

600

800 1000 Asset 2 norm

1200

1400

1600

1800

5 0 -5

200

400

600

800 diff

1000

1200

1400

1600

1800

5 0 -5

200

400

600

800 Density

1000

1200

1400

1600

1800

1 0.5 0 -4

-3

-2

-1

Figure 4. Percentile Triggers.

In the figures above red and green lines are 10% and 90% percentile, blue line is median. So we could trade when our spread is above/ below this value. We sell first asset and buy second when diff is above red line and buy first and sell second when diff is below green lines. We could reverse position at opposite level (when we enter on red then close and open on green and vice versa) but it could be too aggressive (for me at least). I usually close trade when spread return to its mean on zero line. But what percentile or standard deviation multiplier should we use to maximize profit ? First we should check if spread is tradable in the meaning of bid/ask spread and transaction costs. Potential profit should be greater than costs (as always in trading - pretty simple ;). When we use small multiplier of std. dev or big percentile well have more trade but smaller, when we use bigger multiplier or smaller percentile well have less trade but bigger. So we should calculate optimal levels. It could be done by counting trade level crossing (how many trade) and calculate profit as proportional to that level (bigger levels = bigger profits on trade) .On fig. below we see count of level crossing as a function of standard deviation.

Count of trade level crossing 250

200

Count of level cross

150

100

50

0.5

1.5

2 multipler of std. dev

2.5

3.5

Figure 5. Trade level crossing So we could estimate optimal level by multiply level crossing count by that level

Profit profile

80 70 60 50
Profit

40 30 20 10 0 -10 0 0.5 1 1.5 2

2.5

3.5

Figure 6. Profit profile As we see max profit we get about 1.2*. To calculate real profit we should use market prices and real costs, but max profit should be about 1.2* (for this example ).

In overall, the distance approach purely exploits the statistical relationship of a pair, at a price level. As the approach is economic model-free, it has the advantage of not being exposed to model mis-specification and mis-estimation. Cointegration

Cointegration method is little more complicated and one need some statistical background to understand it fully, but it could be use without understanding all theory behind it. Cointegration is an econometric property of time series variables. If two or more series are themselves non-stationary, but a linear combination of them is stationary, then the series are said to be cointegrated v. Stationary means (in our example) that probability distribution does not change when shifted in time. Stationary time series are mean reversing as we need under pair trading.

The steps involved are as follows: 1. Identify stock pairs that could potentially be cointegrated. This process scan be based on the stock fundamentals or alternately on a pure statistical approach based on historical data. Our preferred approach is to make the stock pair guesses using fundamental information. 2. Once the potential pairs are identified, we verify the proposed hypothesis that the stock pairs are indeed cointegrated based on statistical evidence from historical data. This involves determining the cointegration coefficient and examining the spread time series to ensure that it is stationary and mean reverting. 3. We then examine the cointegrated pairs to determine the delta. A feasible delta that can be traded on will be substantially greater than the slippage encountered due to the bid-ask spreads in the stocks. We also indicate methods to compute holding periods. Vidyamurthy (2004) i The model that is most commonly assumed for stock price movement is called a log-normal process; that is, the logarithm of the stock price is assumed to exhibit a random walk. This have some statistical implications (to be found in statistic books). So in first step we calculate log of prices and create two series {log(PAt} and {log(PBt}. Then we try to find linear correlation between this series in mathematical sense: log( ) log( )= + [4]

left side is linear combination of log time series, right hand side represents the residual series and is expressed as sum of two components is equilibrium value and is time series with zero mean. When series is mean reverting we expect that it will oscillate about the

equilibrium. So we can say that spread will oscillate about its mean series.

as described by

time

We have a model, now we should estimate this relationship (thats mean coefficient and equilibrium value).

cointegration

It could be done in different ways, we will use linear regression. So we use some software e.g. Excel (see description of my file below) and calculate this coefficient and residuals. For our AUDUSD NZDUSD (90 days from about March 16 2009) example it could look like this: =1.0481 =0.2525 R2=0.9306

Rsquere say us about how good this linear relationship describe log price difference behavior. There are of course also some statistic like t or f statistic. So we have model that describe linear combination between AUDUSD and NZDUSD log( Where ) 1.0481 log( ) = 0.2525 +

is price of AUDNZD and .

is price of NZDUSD at time t.

Now we can calculate residual series = log( so: = log(

) log(

[5]

) 1.0481 log(

) 0.2525

Residuals plot 0.03

0.02

0.01

-0.01

-0.02

-0.03

-0.04

10

20

30

40

50

60

70

80

90

100

Figure 7. Residuals series plot.

Now we should perform some test if this series are stationary e.g. Dickey-Fuller or PhilipsPerron test. Our series passed it (or if we dont know how, we just dont do it, if we believe that our series is mean reversing). We could use this residual series in similar way as we used earlier series of difference . We could set some trade levels and calculate how many times series cross zero line (mean value of spread) and profit.

Asset 1 log 0 -0.2 -0.4 0 -0.4 -0.6 -0.8 10 20 30 40 50 Asset 2 log 60 70 80 90 100

10

20

30

40

50 Asset log diff

60

70

80

90

100

0.4 0.2 0

10

20

30

40

50 residuals

60

70

80

90

100

0.05 0 -0.05

10

20

30

40

50 Density residuals

60

70

80

90

100

40 20 0 -0.06

-0.04

-0.02

0.02

0.04

0.06

Figure 8. Residuals plot and levels

Profit profile

4 3.5 3 2.5 2 1.5 1 0.5 0 -0.5 0 0.5 1 1.5 2

2.5

3.5

Figure 9. Profit profile In this example max profit is about 0.98*. So, we first calculate regression, than we set trade levels and trade crossing it by residuals. Residuals we calculate using equation [5]. Parameters of this models always change in time. As longer we are in trade its more risky. There is phenomena called mean drift - mean are drifting with time in markets and its not good for us because we trade reverse to this mean. Excel file

Excel file contains some spreadsheets one can use to test pair and find level to trade. In first and second are pair data (in example I use Close prices of USDCHF and EURUSD ). In next CALC 1 are formulas for normalized difference calculating. In CALC 2 are formulas for difference of normalized prices and in Regr are cointegration model. So You could use it to estimate model and trade it. In real trading system I use trading rules as above and some another gadgets like trailing stops (profit trailing) and emergency stop loss. Size of both legs of trade should be calculated using market neutrality and money management rules. Enjoy. Andrzej (Andrew) Endler

[email protected] Vidyamurthy, G. (2004) Pairs Trading, Quantitative Methods and Analysis, John Wiley & Sons, Canada. Binh Do _ Robert Faff Kais Hamza (2006) A New Approach to Modeling and Estimation for Pairs Trading
iii ii i

Gatev, E., G., Goetzmann, W. and Rouwenhorst, K. (1999) Pairs Trading: Performance of a Relative Value Arbitrage Rule, Unpublished Working Paper, Yale School of Management. Nath, P. (2003) High Frequency Pairs Trading with U.S Treasury Securities: Risks and Rewards for Hedge Funds, Working Paper, London Business School.
Wikipedia

iv