MarketTrak's Forecast of the SP500
Notices/Comments: Welcome back!
Our 1-day network model moved to cash for trading on Wednesday. All three swing trade models remain long.
If this page does not look current, try refreshing it by pressing the CTRL and F5 keys.
Visit our Twitter page for early forecast updates and system information. A Twitter account is not needed.
Forecast Models: There are four neural network models and three swing trade models. The models are optimized to maximize the total return over the last two months. All models are experimental and are presented here for information purposes only. Their accuracy, which varies considerably over time, is usually greater than 72 percent based on a 90 day average. Additional details about the neural network model are provided here.
All four neural network models forecast the SP500 and differ only in their forward slope length in days. Three of the four are long-term models and the fourth is a 1-day model. The interpretation of the long-term forecasts needs some discussion. The three long-term models ANO values and ANO trend strength arrows are shown in the Summary above. These forecasts are a prediction of the direction of the forward slope and therefore there will be a delay in the expected change in index prices. The delay is about one-half of the forecast horizon days away. This means that if the 10-day model goes negative today, we would expect the market to turn down in about 5 days. Click here for details on long-term slopes.
There are three swing trade models. One for the SP500, one for the DJIA, and another for the NASDAQ 100. The results of the forecast analysis of these three are shown above. Currently, these models will only be either long or in cash. This will change in the future. The swing trade model is quite different from the neural network model used to make the four SP500 network forecasts. The swing trade model is based on a decision table paradigm whereas the network model is based on a neural network architecture. A genetic programming algorithm is used to create a neural network. An evolutionary programming algorithm is used to create a decision table. The network model is a numerical calculation while the decision table is more logical than numerical. The swing trade model trades less frequently than a network model and its signal may lag the market by a day or two. The network model will anticipate the daily change in the market direction. The network model will do better in a volatile market while the swing trade model will do better in a trending market.
Click here for a roadmap of this page. Click here for performance data. Click here for our message board.
Long-term models explained Follow us on Twitter
Computing the Neural Network Forecasts:
We use an advanced neural network model and
a genetic programming learning algorithm to create the neural networks used to forecast the direction of the SP500.
The output of a neural network is a numerical
value that ranges from minus one (-1.0) to plus one (+1.0), with -1.0 being a very
strong indication of a negative slope over the forecast days
and +1.0 being a very strong indication of a positive slope over the forecast days.
For the 1-day model, the trading position we show in the Summary above is determined by computing
an average network output value (ANO) from the individual networks and applying this average value
to the setpoints which are also shown above. When this ANO rises above the long setpoint, a
long position is given. When the ANO falls below the short setpoint, a short
position is given. A cash position is indicated when the ANO lies between these
two points. For the three long-term models, the ANO values are computed and shown in the Summary but
position setpoints are not defined. A five percent trailing stop is given to reduce risk when either a
long or short position is indicated.
Model Results: Trading results for the 1-day network model covering the last 165 trading days are shown in the two charts below. The vertical grid lines in these charts are 10 trading days apart.
The different color regions in the first chart correspond to the three possible trading positions. The second chart shows the trading positions as determined by the forecast model.
Performance Data: Performance of the forecast model is shown in the table below. The results are given for three time periods. The first column of data shows results since the beginning of this year (Y-T-D). The second column of data shows results for the last 165 trading days, and the third column shows results for the last 45 trading days or about the last two months. The drawdown is the maximum drop in any position's value during the total days shown. The table below and the charts above show blind results after model version date, which is given above in the Summary. Before that date, much of the data shown was used in training the networks and therefore the predictions indicated may not be representative of future behavior of the model. Page Top
Model Output Data: The table below shows the results of calculations using our neural network forecast model. The data in the table were computed by averaging the results of more than 200 networks trained on 7800 days of stock market data.
Disclaimer: Our models are experimental and can change at any time. You may use our forecast as long as you totally agree with the following terms and conditions: Our mission is to make our forecast the best available, but we do not guarantee its accuracy or profitability. We provide the forecast on an as is and as available basis without any warranty of any kind. Specifically, MarketTrak disclaims any and all warranties, expressed or implied, including without limitation warranties of merchantability, profitability, and fitness for a particular purpose. In no event shall you hold MarketTrak liable for any direct, indirect, incidental, punitive, or consequential damages of any kind whatsoever. You bear full responsibility for your own investment research and investment decisions. Nothing in our forecasts should be interpreted as a recommendation or solicitation to buy or sell any security, or to take any specific action. The opinions or comments expressed may change without notice. You determine if and how the forecast is used in your investments. You recognize that past performance does not indicate future results. Click here for a complete disclosure of our disclaimer. Page Top