predictedyi: predicted value i-th data point / observation actualyi: true value for the i-th data point / observation
n: number of data points / observations
Forecast Bias
Measures the systematic deviation of forecasted values from the actual values.
Bias = ( ∑
(predictedyi - actualyi) )
/ n
- positive bias is an overestimation
- negative bias is an underestimation
- Advantages:insights into the accuracy and consistency of forecasts
- Disadvantages: ignores the timing and direction of errors
Forecast Interval Coverage (FIC)
metric used to assess the accuracy and reliability of time series forecasting models
FIC = ((number of actual values within the forecasted intervals) / n) x 100%
- Advantages: simplicity and interpretability
- Disadvantages: does not consider the width of the intervals or the distribution of the forecast errors
Prediction Direction Accuracy (PDA)
Measures the percentage of correct directional predictions
PDA = (( ∑
(Prediction Directioni = Actual Directioni) )
/ n ) x 100%
- Advantages: straightforward and intuitive measure of the model’s ability to predict the direction of future values
- Disadvantages: does not capture the magnitude or precision of predictions