The objectives of this paper are to implement “structural changes detection procedure” into pairs trad-ing algorithm, and to show that the suggested way of pairs trading over-performs the existing model. Differ-ences between two algorithms are rigorously studied through simulations and empirical studies.
Structural changes in pairs trading are defined in two ways. First, changes in cointegrating factors can be defined as structural changes in pairs trading scheme. This kind of changes can be found by adopting An-drews and Ploberger(1994)’s methodology, called ExpF. Broken cointegration relationship can be defined as the second type of structural changes. It can be easily detected by Dickey and Fuller(1981)’s unit root test.
Performances of pairs trading will be analyzed in four dimensions ?? (1) frequency of structural changes, (2) size of structural changes, (3) speed of mean reversion and (4) consecutive days of broken coin-tegration. Conclusions from simulation studies are as thus. The suggested pairs trading algorithm tends to perform better than the algorithm previous defined, when there are more and larger structural changes, and when spread follows quicker mean reverting process. If spread are piecewise-cointegrated, then longer consecutive days of broken cointegration will broaden the performances of two algorithm.
The results from empirical studies are also consistent with the simulation studies. Suggested algorithm always over-performs existing algorithm with regards to risk. More returns, however, cannot be guaranteed by implementing structural changes detection procedure. Even though total profit from suggested way is higher than the return from existing way, difference between two results is statistically insignificant.
Considering that pairs trading is seeking absolute profit rather than relative profit, minimization of loss occurrence probability is important, so that it can be considered as quasi-arbitrage. In this context, suggested way of pairs trading is a much better algorithm than the existing one, since the cumulative Profit/Loss has significant upward-slope with small variance.
The objectives of this paper are to implement “structural changes detection procedure” into pairs trad-ing algorithm, and to show that the suggested way of pairs trading over-performs the existing model. Differ-ences between two algorithms are rigorously studied through simulations and empirical studies.
Structural changes in pairs trading are defined in two ways. First, changes in cointegrating factors can be defined as structural changes in pairs trading scheme. This kind of changes can be found by adopting An-drews and Ploberger(1994)’s methodology, called ExpF. Broken cointegration relationship can be defined as the second type of structural changes. It can be easily detected by Dickey and Fuller(1981)’s unit root test.
Performances of pairs trading will be analyzed in four dimensions ?? (1) frequency of structural changes, (2) size of structural changes, (3) speed of mean reversion and (4) consecutive days of broken coin-tegration. Conclusions from simulation studies are as thus. The suggested pairs trading algorithm tends to perform better than the algorithm previous defined, when there are more and larger structural changes, and when spread follows quicker mean reverting process. If spread are piecewise-cointegrated, then longer consecutive days of broken cointegration will broaden the performances of two algorithm.
The results from empirical studies are also consistent with the simulation studies. Suggested algorithm always over-performs existing algorithm with regards to risk. More returns, however, cannot be guaranteed by implementing structural changes detection procedure. Even though total profit from suggested way is higher than the return from existing way, difference between two results is statistically insignificant.
Considering that pairs trading is seeking absolute profit rather than relative profit, minimization of loss occurrence probability is important, so that it can be considered as quasi-arbitrage. In this context, suggested way of pairs trading is a much better algorithm than the existing one, since the cumulative Profit/Loss has significant upward-slope with small variance.
