Co-integration and Pairs Trading

The co-integration is an important statistical concept behind the statistical arbitrage strategy named “Pairs Trading”. While projecting a stock price with time series models is by all means difficult, it is technically feasible to find a pair of (or even a portfolio of) stocks sharing the common trend such that a linear combination of two series is stationary, which is so-called co-integration. The underlying logic of Pairs Trading is to monitor movements of co-integrated stocks and to look for trading opportunities when the divergence presents. Under the mean-reversion assumption, the stock price would tend to move back to the long-term equilibrium. As a result, the spread between two co-integrated stock prices would eventually converge. Furthermore, given the stationarity of the spread between co-integrated stocks, it becomes possible to forecast such spread with time series models.

Below shows a R utility function helping to identify pairwise co-integrations based upon the Johansen Test out of a arbitrary number of stock prices provided in a list of tickers.

For instance, based on a starting date on 2010/01/01 and a list of tickers for major US banks, we are able to identify 23 pairs of co-integrated stock prices out of 78 pairwise combinations. It is interesting to see that stock prices of two regional players, e.g. Fifth Third and M&T, are highly co-integrated, as visualized in the chart below.

pkgs <- list("quantmod", "doParallel", "foreach", "urca")
lapply(pkgs, require, character.only = T)
registerDoParallel(cores = 4)

jtest <- function(t1, t2) {
  start <- sd
  getSymbols(t1, from = start)
  getSymbols(t2, from = start)
  j <- summary(ca.jo(cbind(get(t1)[, 6], get(t2)[, 6])))
  r <- data.frame(stock1 = t1, stock2 = t2, stat = j@teststat[2])
  r[, c("pct10", "pct5", "pct1")] <- j@cval[2, ]
  return(r)
}

pair <- function(lst) {
  d2 <- data.frame(t(combn(lst, 2)))
  stat <- foreach(i = 1:nrow(d2), .combine = rbind) %dopar% jtest(as.character(d2[i, 1]), as.character(d2[i, 2]))
  stat <- stat[order(-stat$stat), ]
  # THE PIECE GENERATING * CAN'T BE DISPLAYED PROPERLY IN WORDPRESS 
  rownames(stat) <- NULL
  return(stat)
}

sd <- "2010-01-01"
tickers <- c("FITB", "BBT", "MTB", "STI", "PNC", "HBAN", "CMA", "USB", "KEY", "JPM", "C", "BAC", "WFC")
pair(tickers)

   stock1 stock2      stat pct10 pct5  pct1 coint
1     STI    JPM 27.207462 12.91 14.9 19.19  ***
2    FITB    MTB 21.514142 12.91 14.9 19.19  ***
3     MTB    KEY 20.760885 12.91 14.9 19.19  ***
4    HBAN    KEY 19.247719 12.91 14.9 19.19  ***
5       C    BAC 18.573168 12.91 14.9 19.19   **
6    HBAN    JPM 18.019051 12.91 14.9 19.19   **
7    FITB    BAC 17.490536 12.91 14.9 19.19   **
8     PNC   HBAN 16.959451 12.91 14.9 19.19   **
9    FITB    BBT 16.727097 12.91 14.9 19.19   **
10    MTB   HBAN 15.852456 12.91 14.9 19.19   **
11    PNC    JPM 15.822610 12.91 14.9 19.19   **
12    CMA    BAC 15.685086 12.91 14.9 19.19   **
13   HBAN    BAC 15.446149 12.91 14.9 19.19   **
14    BBT    MTB 15.256334 12.91 14.9 19.19   **
15    MTB    JPM 15.178646 12.91 14.9 19.19   **
16    BBT   HBAN 14.808770 12.91 14.9 19.19    *
17    KEY    BAC 14.576440 12.91 14.9 19.19    *
18   FITB    JPM 14.272424 12.91 14.9 19.19    *
19    STI    BAC 14.253971 12.91 14.9 19.19    *
20   FITB    PNC 14.215647 12.91 14.9 19.19    *
21    MTB    BAC 13.891615 12.91 14.9 19.19    *
22    MTB    PNC 13.668863 12.91 14.9 19.19    *
23    KEY    JPM 12.952239 12.91 14.9 19.19    *