Dear Henry,
Tradecision requires a set of input columns to base the neural model training on, and identify price patterns and relationships. All data that as you believe significantly affects the value of the target or represents the market situation should be used as inputs. This includes price data (open, high, low, close and volume), technical indicators (moving averages, lags, MACD, RSI, PercentR, and so on) market indexes (S&P500, Dow Jones Industrial, NASDAQ, and so on), as well as relevant stocks and other information that represents the market conditions.
Inputs are the only kind of knowledge about the market that will be available to a neural model to base its forecasting on. Therefore, selecting the right inputs is the trader’s most important task as the rest of the tasks in Tradecision are automated.
It is worth a mention, that irrelevant or insignificant inputs may deteriorate the model’s performance.
Another important aspect is the right amount of data. If your historical data has a small number of price patterns, the neural network will not have enough information about your market to train correctly. For daily bars, it is recommended that you have price data for at least 4-6 years (1000-1500 bars) for the network training and for at least 6 months for the walk-forward testing.
Too much data may increase a neural model’s training period but will not improve the forecasting quality. Additionally, some old price patterns will simply not be valid in the current market situation. Therefore, to reduce the input dataset you need to remove the oldest data.
Sincerely,
Tradecision Support Team
I want to achieve good forecasting results. What information should I feed into a neural model?
