Using only 1 SMA input instead of the 2 SMAs


This post is the continuation of the previous articles.
Although, it is sensible to use as many indicators as possible for NN input, just for fun, let’s try using only one.
The input is (as previously), the RUT index %distance from its SMA(D).
For instance, if the input is -0.02, that means the RUT is -2% under the SMA(D).

This is the performance of each test:

only SMA(10): winLoseRatios Arithmetic Mean: 53.07%, stdev: 4.27%, avgBarGainPercent mean: 0.08%, stdev: 0.29%, projCAGR: 4.08%***p_test: 302
only SMA(20): winLoseRatios Arithmetic Mean: 53.71%, stdev: 4.05%, avgBarGainPercent mean: 0.10%, stdev: 0.26%, projCAGR: 5.22%***p_test: 324
only SMA(30): winLoseRatios Arithmetic Mean: 51.93%, stdev: 4.82%, avgBarGainPercent mean: 0.08%, stdev: 0.26%, projCAGR: 4.35%***p_test: 303
only SMA(50): winLoseRatios Arithmetic Mean: 52.10%, stdev: 5.15%, avgBarGainPercent mean: 0.06%, stdev: 0.29%, projCAGR: 2.98%***p_test: 315
only SMA(100): winLoseRatios Arithmetic Mean: 52.57%, stdev: 4.84%, avgBarGainPercent mean: 0.13%, stdev: 0.30%, projCAGR: 7.19%***p_test: 312
only SMA(150):winLoseRatios Arithmetic Mean: 52.55%, stdev: 4.56%, avgBarGainPercent mean: 0.12%, stdev: 0.27%, projCAGR: 6.47%***p_test: 360
*only SMA(170):winLoseRatios Arithmetic Mean: 53.80%, stdev: 4.25%, avgBarGainPercent mean: 0.19%, stdev: 0.28%, projCAGR: 10.54%***p_test: 695
only SMA(190): winLoseRatios Arithmetic Mean: 54.62%, stdev: 4.30%, avgBarGainPercent mean: 0.18%, stdev: 0.26%, projCAGR: 9.64%***p_test: 385
only SMA(210): winLoseRatios Arithmetic Mean: 54.28%, stdev: 4.20%, avgBarGainPercent mean: 0.17%, stdev: 0.27%, projCAGR: 9.19%***p_test: 391
only SMA(230): winLoseRatios Arithmetic Mean: 54.52%, stdev: 4.59%, avgBarGainPercent mean: 0.16%, stdev: 0.30%, projCAGR: 8.83% ***p_test: 409

Instead of focusing on the correct direction prediction (winLoseRatios Arithmetic Mean), I prefer to focus on the projCAGR.
For that, the SMA(170) is the winner.

We can see that the NN can deduct and learn little knowledge from short term SMA data.
They are more effective, if they learn the longer term (170 days..210 days) data.

What we learn from this is that if we move to the 2 SMA input case,
it is more important to select the correct longterm day parameter than the shortterm one.

That is sensible, shortTerm is more random, we can gain less from that data.
Most of the knowledge is learnt from the LongTermMA. But the combination of longTerm + shortTerm should improve it a little bit.
As the longTerm is the dominant, we should focus finding a better longTerm;

Based on this test, we suggest that for the longterm parameter we should use 180.
And it doesn’t’ really matter how we pick the short term days parameter. One suggestion is the 20 days.

More tests to follow…


No Responses Yet to “Using only 1 SMA input instead of the 2 SMAs”

  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: