So far we used only closing price. However typical log of stock prices has data about closing prices, volume, open, high and low prices. This is additional information that can be used for prediction also. In this section we will investigate how volume information can improve the forecast. We will use the same model as in last time but now for input in addition to closing prices will be also use volume info. There are several ways how we can input volume information. We will try volume difference, volume itself, log of volume. Also we will try different scaling factors for volume. The results are shown in the table below.

Lead 0-4 for prices
Lead 0-9 for volume
High Filter 0.11 0.01
Low Filter 0.1 0.01
"Error is calculated as SUM(act-predicted)**2 where act, predicted are price differences"
We see that using volume improved prdiction from 3.518 to 2.853 (19%).

Looking at weekly errors we see that our best type is not always the best. On some weeks another models outperform it. If we could switch to the one that is currently doing better we could improve even more our prediction. Obviously we need the way to know how to pick it. This leads to idea of making adaptive system that could evaluate performance and switch to the best model.