veeru,
we have been playing with least square analysis as well and I feel about it as you do. actually we are a team of three people and I will pass on the orthogonal analysis to our statistic guy.
right now it seems that we have profitable tests with sharpe ratios at about 1 or slightly above [over a five year horizon]. we are using automatic pattern recognition for that. it has the advantage that you do not necessarily "see" a trade on the chart. I think this is important because the only reason I can think of why standard tools suck is that it is too obvious. In a to some extent efficient market there is necessarly pairs trading going on.
When I look at pairstrading.com and their recommendations I wonder about these pairs trading in completely different sectors. I always try to understand my trading and I have not got the rationale behind these trades - so far. Let's assume they are working - why are they working? Statistically there is at any point of time a number of accidental outliers. This means that there is necessarily a "stupid" pair, that cannot be traded at all but shows significant correlation in recent history. This is actually the reason why I tried so far to keep my hands away from these trades. But if pairstrading.com find it usefull to trade them, I assume they will have reason to do so - meaning it will work. But why. Does it mean that for whatever reason a big trading book has the two stocks in one basket and buys and sells this basket therefore creating the correlation + possibility to trade? Or is it a game that rather refers to a kind of "overall integrity" of the market than to a pure pairs concept?
One other thing: if you look at two stocks and build correlations on an intraday basis it seems as if there was a constant decline of correlation throughout the day. I wonder why that is so and if it can be of any help for pairstrading. More in detail: take two stocks and use the 9:45 prices of each day and calculate the correlation. Then take the 10:00 prices and do the same and so forth. I found a significant and constant decline in the computer sector in 2002. This in itself would not mean to much but it contrasts with the fact that each stock has its most volatile period in the morning session. There is biggest volume, even bigger than in the closing range, and there is the period where the market makes more range per time than at any oterh point of time. But it seems as if the market was not only volatile but more aware of market integrity. I find this rather interesting.
One argument might be that the market is highly correlated because the biggest nonTradingTime where stocks could not react is just over and now stocks are effected by news in the very same way. Nevertheless, it could mean fo intraday pairs that convergence trades are better placed late during the session (???).
Does any one trade on a purely automatic basis intraday?
peace
we have been playing with least square analysis as well and I feel about it as you do. actually we are a team of three people and I will pass on the orthogonal analysis to our statistic guy.
right now it seems that we have profitable tests with sharpe ratios at about 1 or slightly above [over a five year horizon]. we are using automatic pattern recognition for that. it has the advantage that you do not necessarily "see" a trade on the chart. I think this is important because the only reason I can think of why standard tools suck is that it is too obvious. In a to some extent efficient market there is necessarly pairs trading going on.
When I look at pairstrading.com and their recommendations I wonder about these pairs trading in completely different sectors. I always try to understand my trading and I have not got the rationale behind these trades - so far. Let's assume they are working - why are they working? Statistically there is at any point of time a number of accidental outliers. This means that there is necessarily a "stupid" pair, that cannot be traded at all but shows significant correlation in recent history. This is actually the reason why I tried so far to keep my hands away from these trades. But if pairstrading.com find it usefull to trade them, I assume they will have reason to do so - meaning it will work. But why. Does it mean that for whatever reason a big trading book has the two stocks in one basket and buys and sells this basket therefore creating the correlation + possibility to trade? Or is it a game that rather refers to a kind of "overall integrity" of the market than to a pure pairs concept?
One other thing: if you look at two stocks and build correlations on an intraday basis it seems as if there was a constant decline of correlation throughout the day. I wonder why that is so and if it can be of any help for pairstrading. More in detail: take two stocks and use the 9:45 prices of each day and calculate the correlation. Then take the 10:00 prices and do the same and so forth. I found a significant and constant decline in the computer sector in 2002. This in itself would not mean to much but it contrasts with the fact that each stock has its most volatile period in the morning session. There is biggest volume, even bigger than in the closing range, and there is the period where the market makes more range per time than at any oterh point of time. But it seems as if the market was not only volatile but more aware of market integrity. I find this rather interesting.
One argument might be that the market is highly correlated because the biggest nonTradingTime where stocks could not react is just over and now stocks are effected by news in the very same way. Nevertheless, it could mean fo intraday pairs that convergence trades are better placed late during the session (???).
Does any one trade on a purely automatic basis intraday?
peace
-- statistics is a tricky subject, anyone read 'Fooled by Randomness' by Taleb, it is an excellent book)