The Financial Forecast Center™
Independent. Objective. Accurate
The Financial Forecast Center's primary method of producing all forecasts is a novel, proprietary time delay neural network architecture.
FFC's secondary methodology is multi-channel singular spectrum analysis. We like to think of this as a "second opinion."
In general, a neural network is a computer program that maps a set of input data to specific target data. A time delay neural network is a certain type of neural network that works well with time series data.
FFC has developed and uses a novel, proprietary time delay neural network architecture and training algorithm. Our methodology performs well at mapping real world input data to target data.
FFC's novel, proprietary neural network architecture and training algorithm enables us to answer three very important questions:
2) When? and
3) How much?
The 'what' is what factors, such as interest rates, are important. The 'when' is how far in the past the 'what' is important. And finally, the 'how much' tells us the magnitude the 'what' and 'when' contribute to a forecast.
Singular spectrum analysis, or SSA, is a spectral estimation method used to decompose a time series into sub components such as trend, cyclical components, and noise.
To forecast a particular time series, a set of neural networks for that time series is used to project the past into the future. As an example, the Dow Jones Industrials neural network designed and trained for 1 month forward is used to project prior data into the future, giving a forecast for the DJIA 1 month forward.
As for singular spectrum analysis, a signal subspace is produced and used to generate a 'linear recurrence relation.' Forecasting is facilitated by the continuation of the linear recurrence relation.