Is anybody using the Modeler to trade pairs? Any thoughts on that? I am thinking about to feeding the software with a stock pair ratio and execute orders based on the signals.

New features:
- Agents can now (optionally) also place market orders (besides limit orders). A new "MktOrder" gene has been added for this.
- The Market Depth window now provides a summary table with the number of buy and sell orders grouped by limit and market orders.
- Virtual Market return and volatility data series added for better analysis of virtual market behavior.
- Percentage of agents buying and selling (through the Orderbook data series)

Changes:
- The Orderbook and VM Trades data series now have a parameter to specify market order, limit orders or both order types.
- Standard Edition now support populations of up to 10,000 agents, manual exporting up to 99,999 bars and an unlimited number of models per batch.

Bug fixes:
- Maximizing Market Depth window during model evolution caused exception.
- Market Depth window caused exceptions in case volume stayed 0 for a while.
- Distribution charts with narrow value range or few bins could cause exception.

It will depend on volatility and transaction costs.

52% - 52.5% can be enough as some of our example models with FDA in that range clearly show. For example:

Model                   FDA         Annual Return     Annual Excess Return
FDX                     52.0%           18.7%                      9.1%
LUV                     52.2%            23.9%                    11.6%
Straits Times        52.5%            24.8%                    18.6%
Hang Seng           52.1%            15.9%                      7.2%

These are the results over all model history (ranging from 23 - 29 years for these models) after all transaction costs and excluding dividends.

See the example models on the website.

You can also easily calculate the expected return distribution based on your expected FDA, volatility, and transaction costs, with the Statistical Simulation data series in Adaptive Modeler.

Is anybody achieving excess returns with a model in where the accumulated FDA is just 52% or 52.5% after 15 years of data?

That is the highest FDA I am being able to achieve when using the modeler to trade a Brazilian stock (PETR4).

Changes:
- Trading Signals and Right/Wrong Forecasted Price Changes are now also shown in charts with longer periods (up to 5 years for daily quotes). When a data series can not be shown in a chart because of insufficient space per bar, the data series name above the chart will be shown in gray.
- Position of dialog box for changing data series and performance window parameters is remembered.

Bugs fixed:
- Market Depth window could become very slow or halt in certain cases.

See the example models on the website.

On the download page, you can see which ones were created with the Professional Edition.

The Professional Edition was used to create these particular models (with 10,000 or 20,000 agents) because the results for these securities using the Evaluation Edition (with only 2,000 agents) were poor.  (The DJIA model uses only 2,000 agents. It only requires the Pro Edition because it has more than 20,000 bars).

Hello

I think it would be helpful if someone with access to the professional version to post some highlights on the effects of high agents number in terms of forecasting accuracy. I know it is only evident that higher percentages are to be expected, and the models are subject to different subtle variations in their parameters. However, I do believe it would be interesting (to me at least) to see, all things being equal, the relative effect of an increased agent-population model on the strength of the forecasting accuracy. And whether there is a diminishing marginal benefit in increasing the population size (i.e >80,000) relative to the incremental accuracy and time-spent(computational power).

   
I would be delighted to see some examples (example) where the standardized S&P file was tested within some higher upper bounds (for example, in excess of 50,000).

New features and changes

- Sortino ratio added (as a data series and in the Performance Overview)

- Models can now automatically be updated from the command line using an "/Update" switch (opens a model, evolves it until the end of the quote file, saves the model and exits).

- Window tabs can now be reordered by dragging them with the mouse

- Windows now have a context menu for renaming, removing, etc. (right-click on title-bar or tab)

- Some minor changes and bug fixes

SPY price data only goes back to 1993, its inception year.
For the S&P 500 index, data since 1950 is used. This provides a much longer learning period and covers more diverse market behavior. Models need sufficient evolution time to achieve robust performance.

Since SPY price changes are almost perfectly correlated with S&P500 price changes, you can use a S&P500 model to create signals for SPY.

I have run your evaluation edition a number of times using price data on SPY vs the S&P 500 model and cannot produce returns with SPY that are commensurate with S&P 500 returns.  The best SPY models were only just above 50% FDA vs about 54% FDA on S&P data. Can you explain this?  If I were to use the full edition would it be better to make buy/sell decisions of SPY using the S&P 500 model to generate trades?