Different Order Types

If you are working with daily bars and want your Trading Strategy to exit on market close on the same day you entered a trade, you can change the order type from the default of market to Market on Close (current).  This order type exits at the closing price of the last completed bar. The exit price is the closing price plus slippage.  ChoosingMarket on Close (current)will prevent you from holding a position overnight. *

There’s also a Market on Close (next bar) order that allows you to exit at the closing of the next bar. The exit price is the closing price plus slippage.

*Note that both market on close orders work for backtesting, but in realtime trading they still fire at the end of the bar (and not before the bar ends).  In order to get out before the end of the day in real time, you would have to place a manual trade.   Otherwise, the order would be filled at the next day’s open.

These order types may also be used with intraday charts, but will exit upon completion of a bar or the next bar rather than waiting for the end of day.

In the example above, the close was used as an entry condition, but other conditions may be substituted.  (Since the close of a bar is always present it makes the condition true, which then triggers an entry or exit.)

Different Prediction Outputs 

The Output tab in the Prediction Wizard defaults to the Percent Change in Open.  We set that default based on our experience that it is better to predict a normalized value over time such as Percent Change rather than predicting a price directly.  Even if the prices in your training history have doubled in value over several years of daily training data, the percent change from one day to the next does not usually vary that much.  Therefore, if you predict a percent change your prediction model should be more accurate.

However, it sometimes pays to experiment and I took advantage of the Prediction Wizard’s ability to choose a different output.  You can choose any of the other options in the drop down list or you can specify any other data stream on your chart by selecting Other Data/Indicators.

For this example, I chose the Optimal Buy/Sell Hold based on Open because I wasn’t as much interested in predicting the amount of the change, but whether the price was approaching a peak or a valley.  The output produces a 1 when prices are going upwards towards a peak or a -1 when prices are going downwards towards a price valley.  When values are 0, it signals a hold.  (See the help topic Neural Network – Output Discussion for details.)

The inputs for the model are the Relative Strength indicator from the Price Momentum Category, and the Money Flow Index, which is a volume weighted form of RSI that uses an Average of Open, High, Low, and Close instead of just one price stream.  The closer the Index is to 100, the more positive the money flow, while values near 0 indicate a negative money flow.

When I fed those inputs to the net, the in-sample signals were not very frequent because the market was in trend mode.  The out-of-sample period was more choppy but the model was smart enough to generate signals on the more frequent peaks and valleys.

Position Sizing

Now that I had a prediction model that performed well in the out-of-sample period, I wanted to see what the position sizing option in the Power User version could add to the model.  I created a Trading Strategy based on the neural network entry and exit conditions and I locked the rules.  In the Trading Strategy parameters on the Sizing tab, I chose a fixed size of 10000 units, Fixed leverage, Kelly formula, and Optimal f.  (These different methods are described in detail in the Position Sizing Methods help topic.)  I let the optimizer choose the most appropriate sizing method.

The optimizer chose the Kelly formula as the sizing method.  The net profit for the out-of-sample period rose to $14,818.21 compared to $273.00 for the original prediction.

The chart is available from www.ward.net in the New and Updated Examples section.