Why does FLYR Hospitality track both MAPE and sMAPE?

What is MAPE?

MAPE: Mean Absolute Percentage Error

MAPE's formula: |actual − forecast| / actual

Say you're forecasting room nights on a slow Tuesday where actual demand is 2 rooms, and your forecast was 4 rooms. You're off by 2 rooms. But because you're dividing by the actual (2), the error reads as 100%.

Now say on a busy Saturday, actual demand is 100 rooms and your forecast was 98. You're off by 2 rooms again, but the error reads as 2%.

Same size miss (2 rooms) in both cases, but MAPE says one is a disaster and the other is basically perfect. That's the flaw: when the actual number is small, dividing by it inflates the percentage, even if the underlying miss is trivial. A single slow Tuesday can wreck your average error rate and make the model look far worse than it's actually performing.

What is sMAPE?

sMAPE: Symmetric Mean Absolute Percentage Error

sMAPE's formula: |actual − forecast| / ((actual + forecast) / 2)

Same slow Tuesday: actual 2, forecast 4. Average of the two is 3. Error is 2/3 ≈ 67%. Still shows a miss, but not the exaggerated 100%.

Same busy Saturday: actual 100, forecast 98. Average is 99. Error is 2/99 ≈ 2%. Basically unchanged.

The fix is what's in the denominator. MAPE only looks at the actual value, so small actuals distort everything. sMAPE looks at the average of actual and forecast, which is a steadier number to divide by, so low-demand days don't get an artificially inflated error score.

That's why we track both: MAPE is the simpler, more familiar number, and sMAPE gives a truer read during shoulder season or slow days when actual demand is naturally low.

Want to learn more about how FLYR Hospitality uses MAPE and sMAPE in forecasting and model evaluations? Check out “The Signal” article all about backtesting!