Quant basket trader? I need your help!

Quote:

---

Originally Posted by ;

When you hunt for mean reversion on google you always find a reference to this Ornstein-Uhlenbeck process (or the Vasicek model in finance). You'll also find reference to this discrete equivalent AR(1) autoregressive process. The Ornstein-Uhlenbeck process is formulated as picture Where is the expression, #952# and; 963; are positive parameters. Wt is the Wiener process. I prefer writing this way: picture because today we see dx/dt as a velocity and dW/dt as noise. Wikipedia offers a picture of the development of the process: picture Wherever the beginning...

---

Hello PipMeUp,

Interesting analysis and explanation - Thank you. One thing that I often consider when analyzing the derivatives of price since in subgraph #2 and #4 above from the picture is to also plot the cumulative sums of these two figures to get an idea of how much mean reversion as time passes there is in these measures - and compare the outcome to subgraph #1. I do this because these plots often lie about cumulative mistakes and my eye can not pick up the discrepancy without an alternate visualization.

Here are the accumulative amounts. Interesting indeed to see that the first two remain negative for a long long time... The very long term motion is not perfectly recorded by the purple filter. Nonetheless, it is a decade long cycle!
https://www.forexforum.co.za/attachm...1296876994.png

Quote:

---

Originally Posted by ;

Here are the accumulative amounts. Interesting indeed to understand that the first two remain unfavorable for quite a long time... The very long term movement isn't perfectly captured by the purple filter. Nonetheless, it is a decade cycle! picture

---

Those two are really not bad looking for first order derivatives (speed ) compared to some that I've produced. If you cumulate you amplify the errors over time and this representation of the additive errors may be explained in terms of a long-term tendency or trends in the series that was undifferentiated.

If the underlying series is a sine wave, then your trend estimator (post # 89, high subgraph, purple line) must have cumulative mistakes that remain on the online. However, if your underlying series consists of a single tendency (upward ) then attaining cumulative mistakes on the zero line will imply that half the time/magnitude the tendency estimator efficiently directed the underlying series. If it could do that it would by my definition qualify as the estimator. But that's a fairly high bar.

Quote:

---

Originally Posted by ;

quote The problem is that the Ornstein#8211;Uhlenbeck model is not mimicking the market properly. Their model says that the market is like a spring or a rubber band: the further you stretch it the more it will create a force to come back to its own rest/equilibrum position. Hooke's law. But at the market at some stage the rubber band breaks and market runs away. From the market the spring is not constant! It depends upon the distance. You may view it to the heatmap Post...

---

Yes, since if these eerh... Ornstein-Uhlenbeck guys where on the right path, the oversold/overbought requirements of state a typical stupid RSI would be a great thing (or something to that effect). Which it is not, no matter what period or timeframe it is made by you. Clearly, thats not how the market works. And thats not market plogy works. Its like: OK not its moment to sell/buy! And then everybody sort of forgets what happend before. In general. Which means historical data as an immediate use of future data is: Perhaps not usefull.