Product / Product Information / Neural Nets
|
|
|
A new neural model can be easily created using New Model
Wizard that guides you through the process step by
step.
Four Steps of New Model Wizard
Step 1: Selecting target and lookahead
During this step, you need to define the model target and specify how far into the future you want the model to predict.
Step 2: Selecting training and test data ranges
During this step, you need to define how much data will be used for model preparation. You can also manually allocate a portion of this data to be used for walk-forward testing.
Step 3: Selecting model inputs
During this step, you need to define model inputs. You need
to select those indicators which, as you believe, contain important
data that is vital for predicting the model target. New Model
Wizard allows finding the optimal parameters for the indicators
that will be used by you as inputs.
Step 4: Selecting strategy rules for performance testing
To test a neural model, you need to define a strategy based on the model forecasts (model-based strategy).
Trading strategies produced by New Model Wizard
are based solely on model forecasts. These strategies produce
buy/sell signals that depend only on model forecasts, and, therefore,
allow testing model performance only, and not the money management
or indicator-based rules.
Analyze Model Quality
With the Model Performance Report, you can
analyze the quality of neural model forecasting
using statistical ratios and graphs. You can also identify the
reason for your trading system's poor performance. If the report
ratios are good, you should rather change your strategy settings.
If the report ratios are poor, you should improve your neural
model.
The Model Performance Report provides separate figures for the
training and walk-forward testing periods.
The figures for walk-forward testing are much more important
since they represent the model quality you will achieve when
you use the model with new data. Usually, model performance
during the training period is better than during the testing
one. A good model should have good performance during
both periods.
Prevent Curve-Fitting
Curve-fitting is a dangerous shortcoming caused by a
neural network’s over-training. When a network
is over-trained, it doesn't "learn" price patterns or develop
the ability to generalize, but simply memorizes historical price
data or, in other words, fits its output to the price curve
without learning any internal dependencies.
An over-trained network can produce good results with a training set, but performs unexpectedly badly when used with new data.
Tradecision employs a special algorithm to identify
a clear trend of over-training. It monitors the validation
error dynamics and training progress from different points of
view.
|
