Is there a light at the end of the tunnel? Maybe…


Maybe I found the first method that has a slightly significant predictive power.
NNeurons = 20, ValidationError = 6, Input: only the closeprice% and the volume, (however, inputs can be negative [-1, 1] version.)

Input(:, k) = [p_volumesChange(k), p_volumesChange(k+1), p_volumesChange(k+2), p_volumesChange(k+3), p_volumesChange(k+4),
p_closePricesChange(k), p_closePricesChange(k+1), p_closePricesChange(k+2), p_closePricesChange(k+3), p_closePricesChange(k+4)]

The file: NNTest10.m
****Test: 70.
winLoseRatios Arithmetic Mean: 51.12%, stdev: 2.84%
avgDailyGainPercents Arithmetic Mean: 0.05%, stdev: 0.10%

See the winLoseRatios chart:

70 backtests are quite much (it took 8 hours to test). And after that the winLoseRatio stayed above 51%.
It is not very significant…. specially that stdev is 2.84%. So, there is a very high chance that a new random sample will be under 50% winLoseRatio.
On the other hand, we may not care about the winLose ratio.
It doesn’t matter if we are right only 40% of the time if our wins are significantly better than our loses.
So, the other meauserement is the avgDailyGainPercents.
That is expected to be 0.05% +- 0.10%.
So, it is easy to see that it can be negative easily.

However, considering a 0.05% daily gain, the expected annual gain is 1.0005^260 = 14%.
Without slippage.
So, it cannot be played.
And it is too random.
However, after 70 backtests it looks to be an edge.
It looks like the NN can behave non-randomly.
To be honest, I am not totally convinced yet. But I wanted to report in the blog the first thing that may work.

An improvement would be to put back again the open/high/low prices so, the NN could see all the candlestick values.

Note that the input contains 5 days. When it contained only 1 day, the NN had absolutely no predictive power.
A further test would be to change the number of days in the input, and test the performance of the NN.


No Responses Yet to “Is there a light at the end of the tunnel? Maybe…”

  1. Leave a Comment

Leave a Reply

Fill in your details below or click an icon to log in: Logo

You are commenting using your account. Log Out /  Change )

Google+ photo

You are commenting using your Google+ account. Log Out /  Change )

Twitter picture

You are commenting using your Twitter account. Log Out /  Change )

Facebook photo

You are commenting using your Facebook account. Log Out /  Change )


Connecting to %s

%d bloggers like this: