Worked calculation
Calculate forecast error.
Find the error for one forecast, repeat it for every pair, then calculate the mean. Use the CSV to reproduce the arithmetic.
Synthetic example. All 5 pairs were made up for this exercise. Sample prepared .
Step 01 / One pair
Find the gap, ignoring its sign.
The forecast and example outcome are difficulty changes in percent. Subtract the forecast from the example outcome and take the absolute value. Ignore any minus sign. The error is in percentage points (pp).
- Sample 01: forecast
- 2.0%
- Example outcome
- 3.2%
Absolute error
|3.2 − 2.0| = 1.2 pp
Step 02 / The whole sample
Calculate the error for every pair.
Repeat the same subtraction for each row. Taking the absolute value keeps a forecast above its outcome from cancelling one below it.
| Sample | Forecast (%) | Example outcome (%) | Absolute error (pp) |
|---|---|---|---|
| Sample 01 | 2.0 | 3.2 | 1.2 |
| Sample 02 | -1.0 | -0.5 | 0.5 |
| Sample 03 | 4.0 | 2.8 | 1.2 |
| Sample 04 | 0.5 | 1.0 | 0.5 |
| Sample 05 | 3.0 | 4.6 | 1.6 |
Step 03 / Average the errors
Calculate mean absolute error.
Add the absolute errors and divide by the number of pairs. Mean absolute error (MAE) describes the average distance in this sample.
MAE in this synthetic sample
1.00 pp
(1.2 + 0.5 + 1.2 + 0.5 + 1.6) ÷ 5 = 5.0 ÷ 5 = 1.00 pp
pp = percentage points. The values were made up, so this result does not measure any forecaster’s accuracy.
Check the numbers in the CSVWhat the result can tell you.
A point forecast is only part of the evidence. Compare a forecast recorded in advance with the observed outcome, using the same definition for both. Keep missed predictions in the record and check how far each forecast was above or below the outcome.
Show every observation.
The table includes all five made-up forecast/outcome pairs. Misses should be as easy to check as successes.
Keep the units explicit.
The forecast and observed columns are difficulty changes in percent. Their difference is measured in percentage points.
Keep the claim proportional to the evidence.
These invented rows demonstrate the format and arithmetic. They cannot establish forecast accuracy or support a trading decision.
Limits and provenance
The five pairs do not come from a live model, an observation period or a historical source.
A measured report needs forecast timestamps, observation block heights, the model version, the full evaluation period and a baseline for comparison. This sample does not test forecast ranges or performance across market conditions.
