Reversion To the Mean (RTM) Intraday Strategies
- Thread starter Trend Fader
- Start date
Anyone using Data Visualization Techniques? If you do, can you post an image? I am not talking about a simple time/price chart.
The idea is that humans are better then computers in perception, creativity and general knowledge.
I am just curious of what people a are using.
Here is more info:
http://www.dbs.informatik.uni-muenchen.de/~daniel/KDD97.pdf
Example of it:
http://www.ted.com/index.php/talks/hans_rosling_shows_the_best_stats_you_ve_ever_seen.html
I know this is not RTM but I think you will enjoy it. If not just ignore.
The idea is that humans are better then computers in perception, creativity and general knowledge.
I am just curious of what people a are using.
Here is more info:
http://www.dbs.informatik.uni-muenchen.de/~daniel/KDD97.pdf
Example of it:
http://www.ted.com/index.php/talks/hans_rosling_shows_the_best_stats_you_ve_ever_seen.html
I know this is not RTM but I think you will enjoy it. If not just ignore.
Since the concept of cubic splines came up I wanted to get thoughts regarding this, I understanding Mr. Jurik has a monetary incentive for pushing his method:
A favorite method among engineers for smoothing time series data is to fit the data points with a polynomial (eq, a parabolic or cubic spline). An efficient design of this type is a class known as Savitzy-Golay filters. The chart below compares JMA to a cubic-spline (3rd order) Savitzy-Golay filter, whose parameter settings were chosen top make it perform as close to JMA as possible. Note how smoothly JMA glides through regions of trading congestion. In contrast, the S-G filter is quite jagged. Clearly JMA is, once again, the winner.
Love TED.com, some awesome presentations on it. Would really like to attend one in the future.
A favorite method among engineers for smoothing time series data is to fit the data points with a polynomial (eq, a parabolic or cubic spline). An efficient design of this type is a class known as Savitzy-Golay filters. The chart below compares JMA to a cubic-spline (3rd order) Savitzy-Golay filter, whose parameter settings were chosen top make it perform as close to JMA as possible. Note how smoothly JMA glides through regions of trading congestion. In contrast, the S-G filter is quite jagged. Clearly JMA is, once again, the winner.
Love TED.com, some awesome presentations on it. Would really like to attend one in the future.
Quote from knocks420:
Since the concept of cubic splines came up I wanted to get thoughts regarding this, I understanding Mr. Jurik has a monetary incentive for pushing his method:
A favorite method among engineers for smoothing time series data is to fit the data points with a polynomial (eq, a parabolic or cubic spline). An efficient design of this type is a class known as Savitzy-Golay filters. The chart below compares JMA to a cubic-spline (3rd order) Savitzy-Golay filter, whose parameter settings were chosen top make it perform as close to JMA as possible. Note how smoothly JMA glides through regions of trading congestion. In contrast, the S-G filter is quite jagged. Clearly JMA is, once again, the winner.
Love TED.com, some awesome presentations on it. Would really like to attend one in the future.
If you are using a smoothing spline as I mentioned earlier in the 1st example posted, it is a simple matter to adjust for smoothness as there is a coefficient that can be adjusted for that. Note that smoothing spline and cubic spline are two different methods, although both also have dependencies on parameters like number of knots, sample spacing, etc..
You need to give a more objective measure of what constitutes a better fit for purposes of your discussion in order to draw any type of conclusions.
Lo,
Attached a PDF of charts I threw together very hastily based on your last post. I emphasize hastily so please let me know if there are errors. EQ is just a snapshot of earlier posted curve.
Ultimate arbiter of effectiveness of a technique, MA, indicator, whatever is profitability. If someone has the time, can you post an equity curve of same strategy applied utilizing Spline (or other technique) and WMA. Unoptimized is probably best approach.
This is somewhat selfish request as if the marginal difference in profitability between the two is small (Look Ma, 2 extra bps!) I'll try to focus attention on other ideas then to try to decipher/code for cubic spline for my software...
Attached a PDF of charts I threw together very hastily based on your last post. I emphasize hastily so please let me know if there are errors. EQ is just a snapshot of earlier posted curve.
Ultimate arbiter of effectiveness of a technique, MA, indicator, whatever is profitability. If someone has the time, can you post an equity curve of same strategy applied utilizing Spline (or other technique) and WMA. Unoptimized is probably best approach.
This is somewhat selfish request as if the marginal difference in profitability between the two is small (Look Ma, 2 extra bps!) I'll try to focus attention on other ideas then to try to decipher/code for cubic spline for my software...
Consistent with what I said in a previous post, the residuals go out of wack when the volatility peaks. The vol is what breaks your stationarity in your model and ruins your predictions.
So the question you have to ask yourself is: Does profitability drop when the volatility profile shifts substantially?
If it does, you need to hedge out the risk from volatility (via a pair), or adjust how you seek alpha as a function of some type of volatility forecast, like GARCH.
You can bracket your volatilities into arbitrary ranges [e.g., 10-30% = 1, 30-50% = 2, etc.) and assign them ordinal values and find out the expectation of profitability conditioned on the ordinals associated with ranges of volatility. i.e.:
E[Profit | Ordinal = 1 ] = something
E[Profit | Ordinal = 2 ] = something
and so on...
Or do a regression. If the regression parameter associated with the ordinal value corresponding to bracketed volatility shows the expectation drops, then simply use the ordinal value as an input into your system. Problem solved, non-stationarity mostly avoided, and better equity curve ahead.
)
Thought I'd bump this thread to ask a quick q.
How many of you have a method which automatically adjusts to volatility?
Seems to me, if you're using a static number such as a percentage from the mean, you could get destroyed numerous days in a row in periods of high volatility, rather than taking advantage of it.
I'm constantly adjusting manually for now, but have started to work on ways to have the system automatically adjust itself, = wider/ narrower parameters.
Thanks.
How many of you have a method which automatically adjusts to volatility?
Seems to me, if you're using a static number such as a percentage from the mean, you could get destroyed numerous days in a row in periods of high volatility, rather than taking advantage of it.
I'm constantly adjusting manually for now, but have started to work on ways to have the system automatically adjust itself, = wider/ narrower parameters.
Thanks.