Pairs Trading Case Study

METHODOLOGIES
Our implementation of pairs trading has two stages. First stage is pair formation stage. We form pairs over a 12-month period (formation period) and trade them in the next 6-month period (trading period), which is the second stage. The reason for us to separate these two periods is to distinguish back-testing and forward testing in evaluating a trading strategy.
Back-testing refers to use historical data to verify how a trading system would have performed during the specified time window. Though back-testing can provide valuable information to investors in the market, it is often misleading and has been sometimes over focused. To provide a comprehensive evaluation on trading strategy, in addition to back-testing, out-of-sample

And the second trading was open on day 92 while on day 101, the trading position of the two stocks suddenly reversed, in this case, we would use one day to close the position and on the next day to enter an opposite trading position. Then after day 101, we observed that the price ratio between BRCM and EQIX enlarged as the days went on and the position remained open till the end of the trading period. However, we could also observe that EQIX kept going up to catch up BRCM while BRCM dropped from its local highest price and reduced the normalized price difference. This could illustrate that though there may very much fluctuation during each trading interval, the concept of pairs trading is that the pair would eventually go back to their balanced

From the performance test of this pair stocks under “consistent buy-in” procedure, we can see when the initial position opening benchmark is 2 SD or below, loss always exceeds 100%! Even under the strictest opening benchmark, 3SD, the loss still reaches over 50%. A B SD 0.5 SD 1 SD 1.5 SD 2 SD 2.5 SD 3 SD
MSFT FSLR 0.43 -123% -123% -120% -102% -81% -58%
Table 3.3.2 Example of negative return under consistent buy-in procedure.
After a re-evaluation of the correlation of these tow stocks during the trading time, the SD of their normalized price differences have been reaching 1.424593, which is more than three times of the SD previously in the formation period. Notice that the mean prices and SDs calculated from the formation period have been used to calculate the new SD for the trading period, in this case, we can say the price correlation of these two stocks have deviated much away from the standard, and such a significant loss under consistent buy-in procedure makes sense indeed under this condition.
Stop-Loss