Can linear regression analysis really predict the future?

[Tompson]

Yes, great paper.

40 downloads of my last plot but this thread seems to be dying ...

 

[trackstar]

its hard to take it much further. at this point each person must do their own work....

 

[stevegee58]

Quote from Tompson:

Here are 3 distributions for the last 1000 daily bars on SPY for:

close minus open
close minus cubic spline
close minus pivot

The cubic distribution is not normal, I'm not sure why that is. The best curve for a mean reversion trading approach looks to be the close minus pivot.

Maestro, any thoughts on this before I test some trading rules?


Tom

I want to make sure I'm interpreting this correctly. The difference between a day's close price and its pivot for that day is very frequently very small?

 

[Tompson]

Yes, the pivot is simply (C+H+L)/3.

If the high and low are equally spaced around the close then pivot = close.

(C + (C+x) + (C-x))/3 = C

If the high is 4 times further away from the close than the low then the pivot is at:

(C + (C+4x) + (C-x))/3 = C + x

so it takes quite an imbalance to move the pivot.


Tom

 

[Darklingg]

Quote from Tompson:

Here are 3 distributions for the last 1000 daily bars on SPY for:

close minus open
close minus cubic spline
close minus pivot

The cubic distribution is not normal, I'm not sure why that is. The best curve for a mean reversion trading approach looks to be the close minus pivot.

Maestro, any thoughts on this before I test some trading rules?


Tom

Hi,

Here's what I get for EUR/USD using 30min bars.


Btw, I get negative expectancy when trying to replicate your strategy (buying when close below pivot, selling when close above pivot). Did you take commissions into account?

Thanks,

 

[Tompson]

Nice curves - looks like your spline fit is better than mine (MATLAB?).

My test was frictionless but could be reproduced with Direxion funds or similar, given the high number of trades.

However, the equity curve is not great in my view. Something like this could be more promising:

http://cssanalytics.wordpress.com/2...-implied-volatility-vs-historical-volatility/

using IV-HV 75%+


Tom

 

[Darklingg]

Quote from Tompson:

Nice curves - looks like your spline fit is better than mine (MATLAB?).

My test was frictionless but could be reproduced with Direxion funds or similar, given the high number of trades.

However, the equity curve is not great in my view. Something like this could be more promising:

http://cssanalytics.wordpress.com/2...-implied-volatility-vs-historical-volatility/

using IV-HV 75%+


Tom

No, actually I use my own platform.

I made a few changes to the way I compute the interpolation.
The results are much better.

However I still get several 10+ sigma-events.
Any idea how to eliminate them?

I can't get a positive expectancy strategy yet.

 

[Darklingg]

Quote from MAESTRO:

6. The distribution of the actual price fluctuations around a spline predicted point is always Gaussian with the standard deviation that is typically smaller than a price deviation over one time interval used to calculate the center of gravity itself.

This only seems to be true if the underlying distribution is also gaussian.

When I have fat-tails in the distribution, deviations from spline are not gaussian (altough "more" gaussian than the initial distribution).

 

[trackstar]

nice dark.

 

[nfactorial]

Quote from MAESTRO:

Any STABLE distribution has a logic (set of trading rules) that has positive expectations. For example: if the markets were truly Gaussian then the short option strangle positioned at 2 STD would be 100% guarantee winning strategy providing there is a stop loss at 1 STD. You can verify it for yourself.

I don't get this.

By stable you mean stationary I guess?

What do you mean with markets being truly gaussian? Normal distribution of daily returns?

How do you position a strangle at 2 STD? Break-even points should be at current price +/- 2STD?

I'm very interested in this, could anyone explain please?