Example Models

This page shows the performance of the example models and explains how they were created. We also explain how they can be used to view their historical evolution, verify
"out-of-sample"^[2] evolution since their creation, and continue model evolution for ongoing performance evaluation. Finally we explain how these models can be used for trading.

The example models were created on the model creation date shown in the tables below (not to be confused with the historical model start date) and are still evolving and generating signals as of today. Because they were created with the Adaptive Modeler version that was current at that time, they don't make use of improvements brought by later versions.

The tables show the Trading Simulator performance of the models^[3]. The historical results are the results from model start date until model creation date.

It should be noted that the historical model results may seem less impressive than those reported by many other trading software systems that use some form of optimization. However, as explained here, Adaptive Modeler's incremental walk-forward approach (not optimized) prevents overfitting and leads to more resilient models whose historical results are more reliable and more indicative of future results.

The results since creation are the results from model creation date until recently. Since the historical period generally spans several decades while the period since creation is much shorter, relatively less significance should be attributed to the results since creation. Models that have performed well historically, may also sometimes undergo periods of poor performance and may therefore (temporarily) show poor returns since creation date^[4].

More extensive performance information is available inside the models.

How were these models created?

• Security selection
  In general, securities were selected based on availability of sufficient historical data.
  Stock indices were selected among major global stock markets.
  US stocks were selected from a short list of 16 well known stocks that were chosen out of all US stocks with a stock price history dating back to 1985 or earlier and a market capitalization between 5 and 50 billion USD.
   
• Historical data used
  Most models use historical prices going back to 1985 or earlier. Daily closing prices or open, high, low, close bars were used. For some models, volume data was also used. For stock indices, cash prices were used. For commodity ETFs, cash prices were used for historical data before the ETF inception. For other commodities, futures prices were used. Historical prices before model creation date were adjusted for splits and dividends.
   
• Model parameters
  Most models were made with the Evaluation Edition using the default model parameter settings. Models made with the Standard or Professional Edition used 10,000 or 20,000 agents and some other modified settings. The specific parameter settings for all models can be seen inside the models.
   
• Performance calculation
  All returns are unleveraged and after all transaction costs but exclude dividend and interest payments^[8].
  For every model, historical performance is calculated since the start of the Trading Simulator. For the S&P500 index model, this is at model start date. For most other models this is 2,500 bars (about 10 years) after the model start date.
  The risk free rates used for calculating the Sharpe ratio for each model is obtained by taking the average rate during the calculation period of 3-month treasury bills of the relevant country. For commodities, US rates are used. The exact risk free rate values can be seen inside the models (in the Performance Overview).
   

How to verify out-of-sample (walk-forward) performance?

All the downloadable model files are copies made on the model's creation date and were uploaded to the website on that date. The downloadable model files have not been modified since then so they still reflect the situation on creation date. This allows users to observe and verify "out-of-sample" model evolution and performance since model creation date for themselves^[2][5].

To download and open a model in  Adaptive Modeler, follow these steps:

1. Download the corresponding .zip file from the download page.
2. Extract the .zip file to a folder of your choice. It contains a model file (.aam) and a small quote file (.csv) that contains only the last quote row that was read by the model. 
3. Open the model in Adaptive Modeler. When asked for the location of the quote file, specify the folder where you extracted the small quote file (.csv) into.

You will then see the model as it was on creation date. The Performance window will show the same historical results as shown on this webpage. (Note: the S&P 500 model will immediately continue evolving through the rest of the included quote file after opening).

To continue model evolution until the present, follow these steps:

1. Add new quotes to the quote file (use the same formatting as the existing quote row).
2. If the model is paused, press F3 to resume model evolution or press F4 repeatedly to evolve stepwise bar-by-bar.

The model will then continue its evolution by processing the new quotes from the quote file and generate bar-ahead forecasts and trading signals. Once it reaches the end of the quote file, it simply waits for new quotes to be added.

To see the performance since model creation date:

1. Go to the Performance window
2. Click the "Performance Calculation Settings" button
3. For the "Calculation period", select the radio button "Since" and enter the model creation date
4. Click "OK"
    

How to use these models for trading?

To get undelayed forecasts and signals for the future, the Standard or Professional Edition of Adaptive Modeler is needed^[1]. Futures, ETFs or other derivatives could be used to trade indices and commodities. Note that the forecasts and signals in these example model are based on daily closing prices. Therefore, orders could be placed shortly after regular market closing in an extended trading session or at the next day's open.