News for hackers. Stuff that matters.
Re:Why It Might Be a Good Time to Revisit Ray Dalio’s 1937 A... by curiousgal on Thursday July 26, 2018 @05:12PM (#17621184)
Why is that a bad thing?
Re:Why It Might Be a Good Time to Revisit Ray Dalio’s 1937 A... by darawk on Thursday July 26, 2018 @05:30PM (#17621391)
Prices aren't stationary, which means (essentially) that they move around and compound on themselves. Correlation assumes that the two time series being correlated are stationary. If you want to correlate two price series, you do so using returns (e.g. +1%, -2%, ...), which are a stationary series. If you run a correlation on price series, you'll get silly high values like 94%.
Re:Why It Might Be a Good Time to Revisit Ray Dalio’s 1937 A... by eigenvalue on Thursday July 26, 2018 @05:36PM (#17621449)
I'd go further and suggest using changes in the logs of the price over time.
Re:Why It Might Be a Good Time to Revisit Ray Dalio’s 1937 A... by darawk on Thursday July 26, 2018 @05:49PM (#17621576)
My understanding is that using the first difference of the log is just an approximation of % change that is in common usage primarily because it was computationally faster and because it's closed form is smooth. I could be wrong though. Is there some advantage to doing that that i'm unaware of?
Re:Why It Might Be a Good Time to Revisit Ray Dalio’s 1937 A... by merlincorey on Thursday July 26, 2018 @05:30PM (#17621385)
In unsophisticated terms, I think it's basically that you can probably play with nearly any two time series graphs and then manipulate the data such that you find correlations.

Of course, that doesn't mean there actually is any correlation.

Re:Why It Might Be a Good Time to Revisit Ray Dalio’s 1937 A... by darawk on Thursday July 26, 2018 @05:31PM (#17621397)
No, it's a specific issue with price series. The correct way to do it is to correlate the return series.
Re:Why It Might Be a Good Time to Revisit Ray Dalio’s 1937 A... by merlincorey on Thursday July 26, 2018 @05:39PM (#17621479)
I agree that's a bigger fundamental issue, but I still think that if you take two different arbitrary windows from two time series of pricing data and then adjust the scaling of one to "match" the other, one can "find" many of these "correlations", even within the same pair of time series.
Re:Why It Might Be a Good Time to Revisit Ray Dalio’s 1937 A... by darawk on Thursday July 26, 2018 @05:50PM (#17621594)
Ya, you're right. That is an issue as well, though I think it's a second order one here. His correlation likely disappears if he does it on the returns. But spurious correlations, even done mathematically correctly, abound.