Timeline for How do you decide what time frame you're going to use when testing for cointegration?
Current License: CC BY-SA 3.0
12 events
| when toggle format | what | by | license | comment | |
|---|---|---|---|---|---|
| Nov 14, 2016 at 18:36 | history | edited | Richard Hardy | CC BY-SA 3.0 |
added 689 characters in body
|
| Nov 13, 2016 at 22:18 | comment | added | Spencer Smolen | My fear is that I might pick a timeframe, say 5 years, that will return a p-value suggesting that the two timeseries are cointegrated, however they won't "act cointegrated" in the 2 or 3 months months after I run my backtest. Or visa versa, maybe the timeframe of my cointegration test might be too short and suggest that they are not cointegrated, when in fact they are. I know my coint. test vary wildly depending on my timeframe, and considering I'm going to allocate money based on the results of my cointegration test I want to understand why the results vary so much based on time frame | |
| Nov 13, 2016 at 22:12 | comment | added | Spencer Smolen | first of all I want to say thanks for all your help, but I would like to hear other answers. It sounds like you're answer is telling me under what conditions a timeseries will be cointegrated, but my question is more of research related question than a timeseries analysis question. I'm looking for cointegration because I want to be able to trade a pair of stocks that is both (1) cointegrated in my backtest and (2) will likely be cointegrated in the coming months. | |
| Nov 13, 2016 at 16:47 | comment | added | Richard Hardy | @SpencerSmolen, what happened? I see you un-accepted the answer. | |
| Oct 16, 2016 at 20:26 | vote | accept | Spencer Smolen | ||
| Nov 12, 2016 at 20:11 | |||||
| Oct 16, 2016 at 19:57 | comment | added | Richard Hardy | Yes, my data generating process is probably your stochastic process. (I am coming from statistics and time series analysis.) I would not try tailoring the data to the model -- rather the model to the data. But having a model is certainly useful (or indeed instrumental) for testing for cointegration. | |
| Oct 16, 2016 at 19:52 | comment | added | Spencer Smolen | oh i see, when you say data generating process do you mean stochastic process? does this mean that you create a model of the price history and run a test for cointegration on the data that fits the model? | |
| Oct 16, 2016 at 18:11 | comment | added | Richard Hardy | @SpencerSmolen, I am not sure about the equality there, but if the market conditions change, then you could expect cointegration to appear where it was absent or vanish where it was present. | |
| Oct 16, 2016 at 18:02 | comment | added | Spencer Smolen | thank you both! this is great information. Just to clarify, data generating process fixed over time = market conditions fixed? | |
| Oct 16, 2016 at 13:14 | comment | added | Alex C | Let me add a minor anecdote. A friend of mine discovered a profitable trading strategy for ETFs in different countries based on co-integration that would have made great profits in 2008-2009-2010. In 2011 he began trading it and did not make a penny. The big disruptions of the Finl Crisis were over, and his ETFs were no longer diverging and converging in price across time zones as before. As Richard would say it is likely that the data generating process had changed over time. | |
| Oct 16, 2016 at 12:18 | history | edited | Richard Hardy | CC BY-SA 3.0 |
deleted 84 characters in body
|
| Oct 16, 2016 at 10:10 | history | answered | Richard Hardy | CC BY-SA 3.0 |