right, to unbuckle my positions taken on June 30th (no reverse) , according with mean reversion achieved on my own model. Monday 11 : long ATI short FAST long DE short EMR
To all other PTF USER: I have the following issue and looking to see if it only pertains my setup. I notice PTF v301 and v303 are returning different values. Even PTF v301 returns different co-integration and correlation values from two different database - one created 40 days ago and one created today. Ran the 'find pair' function no both database for the same industry and compared the results. They are very different! I was wondering if anyone did some tests with both PTF v301 and v303 and see how the values returned. I upgraded from PTF v301 to v303 and noticed both the co-integration and correlation values of my pairs and it changed dramatically. Needlessly to say it was a confidence bumper. However, after running several tests this morning I was baffled even more. I emailed Jared regarding this issue and I was told that he is not aware of any other people experiencing this same issue. So I ran a couple of tests and lay it out in a spreadsheet for those using PTF v301/v303 to see. I also forwarded this result to Jared and hopefully he will have this sorted out. Observation: 1. v301 and v303 definitely returned different values (even on a fresh database created using the respective version) 2. testing in backward or in forward sequence (alphabetically from a to z and z to a) returned different values for the same pair even in the same version. This should be the same. 3. compared to a database created 40 days ago, created using PTFv301, returns different value from a database (stock groups) created today, also created using PTFv301. (Two tests were performed today - on the old database and on the new one created today.) Thanks - BT Here are the tests: TEST1 PTF v301 with existing database (created when I first purchased the software ~ 40 days ago). The group has the following value: max day in trade = 14 Gather back data for: 3 years TEST2 PTF v301 with new database with default parameters. Old database is deleted using SQL Server Manager Studio. Gather back data for: 3 years TEST3 PTF v303 with new database created - did not use the PTF v301 database. Gather back data for: 3 years TEST4 PTF v303 with the same database as TEST3 but one hour later Gather back data for: 3 years TEST5 PTF v303 with the same database as TEST3 but in backward sequence (starting from the bottom of the list) Gather back data for: 3 years TEST6 PTF v303 with old database (same database as TEST1) max day in trade = 14 Gather back data for: 3 years
To all other PTF USER: I have the following issue and looking to see if it only pertains my setup. I notice PTF v301 and v303 are returning different values. Even PTF v301 returns different co-integration and correlation values from two different database - one created 40 days ago and one created today. Ran the 'find pair' function no both database for the same industry and compared the results. They are very different! I was wondering if anyone did some tests with both PTF v301 and v303 and see how the values returned. I upgraded from PTF v301 to v303 and noticed both the co-integration and correlation values of my pairs and it changed dramatically. Needlessly to say it was a confidence bumper. However, after running several tests this morning I was baffled even more. I emailed Jared regarding this issue and I was told that he is not aware of any other people experiencing this same issue. So I ran a couple of tests and lay it out in a spreadsheet for those using PTF v301/v303 to see. I also forwarded this result to Jared and hopefully he will have this sorted out. Observation: 1. v301 and v303 definitely returned different values (even on a fresh database created using the respective version) 2. testing in backward or in forward sequence (alphabetically from a to z and z to a) returned different values for the same pair even in the same version. This should be the same. 3. compared to a database created 40 days ago, created using PTFv301, returns different value from a database (stock groups) created today, also created using PTFv301. (Two tests were performed today - on the old database and on the new one created today.) Thanks - BT Here are the tests: TEST1 PTF v301 with existing database (created when I first purchased the software ~ 40 days ago). The group has the following value: max day in trade = 14 Gather back data for: 3 years TEST2 PTF v301 with new database with default parameters. Old database is deleted using SQL Server Manager Studio. Gather back data for: 3 years TEST3 PTF v303 with new database created - did not use the PTF v301 database. Gather back data for: 3 years TEST4 PTF v303 with the same database as TEST3 but one hour later Gather back data for: 3 years TEST5 PTF v303 with the same database as TEST3 but in backward sequence (starting from the bottom of the list) Gather back data for: 3 years TEST6 PTF v303 with old database (same database as TEST1) max day in trade = 14 Gather back data for: 3 years
To all other PTF USER: I have the following issue and looking to see if it only pertains my setup. I notice PTF v301 and v303 are returning different values. Even PTF v301 returns different co-integration and correlation values from two different database - one created 40 days ago and one created today. Ran the 'find pair' function no both database for the same industry and compared the results. They are very different! I was wondering if anyone did some tests with both PTF v301 and v303 and see how the values returned. I upgraded from PTF v301 to v303 and noticed both the co-integration and correlation values of my pairs and it changed dramatically. Needlessly to say it was a confidence bumper. However, after running several tests this morning I was baffled even more. I emailed Jared regarding this issue and I was told that he is not aware of any other people experiencing this same issue. So I ran a couple of tests and lay it out in a spreadsheet for those using PTF v301/v303 to see. I also forwarded this result to Jared and hopefully he will have this sorted out. Observation: 1. v301 and v303 definitely returned different values (even on a fresh database created using the respective version) 2. testing in backward or in forward sequence (alphabetically from a to z and z to a) returned different values for the same pair even in the same version. This should be the same. 3. compared to a database created 40 days ago, created using PTFv301, returns different value from a database (stock groups) created today, also created using PTFv301. (Two tests were performed today - on the old database and on the new one created today.) Thanks - BT Here are the tests: TEST1 PTF v301 with existing database (created when I first purchased the software ~ 40 days ago). The group has the following value: max day in trade = 14 Gather back data for: 3 years TEST2 PTF v301 with new database with default parameters. Old database is deleted using SQL Server Manager Studio. Gather back data for: 3 years TEST3 PTF v303 with new database created - did not use the PTF v301 database. Gather back data for: 3 years TEST4 PTF v303 with the same database as TEST3 but one hour later Gather back data for: 3 years TEST5 PTF v303 with the same database as TEST3 but in backward sequence (starting from the bottom of the list) Gather back data for: 3 years TEST6 PTF v303 with old database (same database as TEST1) max day in trade = 14 Gather back data for: 3 years
I noticed v303 calculating differently too. I asked Jared and he said they modified the code to only look at 1 year data in v303 to avoid software from crashing. He thinks this number is better or more informative. I don't know what to make of the whole thing. I am running DFuller tests in matlab as well and it does seem that cointegration moves a lot based on near term data. e.g. SAIA / XTN was a great pair and both CADF and PTF numbers calculated higher COINT. Recently (past couple of weeks) the pair has gone to hell and CADF test says they are not co-integrated any more. PTF v301 agrees and shows 13% cointegration. So it works but if you get in on a pair and it goes to fell and cointegration number falls off do you get out ? or wait to see if comes back. Half life of this pair is like 12 days so do you wait ?
With less than 6 months data S&P Transports will not give stable results. Nor will a single number (which is what? the confidence interval?) convey the tradeability of a cointegrated pair.