Building Neural Network models for short price history

Suppose we want to build Neural Net (NN) model for some financial instrument, suppose it is DAX. with available data since 1987. And we have some verified NN models calculated for DJII (with data available since 1885, almost 135 years of price history already; the NN models were verified in November 2018). We know that having more price history gives more possibilities to verify our models.

The question is: can we use the knowledge gathered from DJII to make a forecast for DAX index with lesser price history? The answer is: yes. With one condition. And the condition is: we assume that DJII and DAX are moved approximately by the same forces. In this case the knowledge gathered from DJII research is useful for DAX. The best way to verify this assumption is simply to draw both these chart together:

The black chart is DJII, while the red is for DAX. As you see, they definitely correlate. This becomes more obvious if we draw these charts in a smaller scale:

It could be if these chars are moved by the same forces. However, the difference presents as well, and we want to catch this difference.

You can do this employing a template technology. Let's do it together, step by step:

Step #1: Download Dow data since 1885.

Step #2: Apply NN models. We have conducted a huge back testing for Dow using many different NN models. The results are stored in Forecast Mill (FM) library:

Here we select several NN models (3 in the example) combining them into one hybrid NN model that covers geo and helio latitude, declination and astrological dignities.

Then click OK to train this new NN model. This model will be applied to create Dow projection line.

Step #3: Save your work with this model for Dow into TS worksheet. Let it be the file Dow_NN_Template.wts . In this worksheet we have stored the information gathered from Dow price history.

Step #4 : Now download the price chart with a shorter price history for the financial instrument we need to forecast. Let it be DAX index.

Step #5: Download your worksheet for Dow AS TEMPLATE. In order to do this, follow menu "File"=>"Run as Template" and choose there your worksheet for Dow (Dow_NN_Template.wts).

You will get the NN model for Dow together with DAX chart.

Now we will work with NN module as usual. As it is now on the back ground, we need to make it visible. To do that, click this button:

The major idea of this approach is: we use the information gathered by Neural Net from analysiing Dow price history and combine this information with the information gathered from DAX price history.

Step #6: To get information from DAX, simply train one more Neural Net using DAX price. Click "Train" button in NN module:

There are two recommendations regarding the training your NN:

a) Do not train NN too long, 5-10 epochs is enough. The program shows the amount of trained epochs here:

Otherwise the information gathered from Dow price will be destroyed.

b) This is optional. You can try to use ALL available DAX price to train your NN. In order to do that, click this "All" button BEFORE training the NN:

The above method may be useful to analyze related financial instruments when one of them has not much price history. You may try to apply it to global indexes compared to regional ones or to pairs "index - stocks".