Monday, February 23, 2009

Qs Linear Regression Crosses

Here's another snippet of the Qs mean reversion signal I've been working on. In much the same manner as the 8/16 moving average crossover this algorithm focuses on the reliability of using the crossover of 2 linear regressions to signal impending reversals of momentum.

In coming posts I'll detail my reasons for favoring linear regression and linear regression channels over other forecasting models. For the near term I'll just mention that linear regressions channels are unique (IMHO) in their ability to capture the trend of both mean reversion and random walk price patterns.

Below is the optimization report for a close range of inputs. What's interesting here is the robustness of the crossover, almost regardless of the settings. I've optimized for max consecutive losers (1), just because risk avoidance is more important to me than maximizing equity output.
Other traders may not be as risk adverse and may choose other input options.
The (9,13,10) model also produces nice returns.

The TS2000i code is very simple and, as with previous MR studies, I'm just looking at the short side for now. I've used a Fixed Bar exit for the study in lieu of a signal reversal. This tactic gains a few more bucks on the equity curve and knocks down the risk exposure time considerably.
The fixed bar exit is (no surprise) 9 days.


Unknown said...

What is a "trend of a random walk price pattern"

bzbtrader said...

Good question. When I refer to the mean reversion trend I'm generally referring to the short term support/resistance channel and the price momentum within that channel. Of course, "short term" is a relative term and depending on your time perspective, may be a single day to several months.
The random walk trend I refer to is the breakout or breakdown in price from a baseline mean reversion channel. I've mentioned before Worden's concept of "kissing the channel good-bye" as one example of such behavior, while the dramatic price swoon of 2001/2002and the subsequent reversal to the upside in 2003 were both longer term examples of random walk trends(IMHO).
Hope that helps.