Question 1

How to detect trend and seasonality in sales data?

Accepted Answer

Trend is detected by computing the Pearson correlation between time index and sales values. Correlation above 0.3 indicates upward trend, below -0.3 indicates downward trend. Seasonality is detected by computing autocorrelation at different lag periods and identifying the lag with the highest autocorrelation above 0.3 threshold. Common seasonal periods are 4 (quarterly), 7 (weekly in daily data), 12 (monthly), and 52 (weekly in yearly data).

Question 2

Which forecasting model should I use for my data?

Accepted Answer

Model selection depends on the time series patterns present. Data with no trend and no seasonality suits Simple Moving Average or Simple Exponential Smoothing. Data with trend but no seasonality suits Holt Linear Trend method. Data with both trend and seasonality suits Holt-Winters method. Data with only seasonality suits Seasonal Naive. The best model is the one with the lowest MAPE on historical data.

Question 3

What is a good MAPE for demand forecasting?

Accepted Answer

MAPE below 10% is excellent, 10-20% is good, 20-30% is acceptable for volatile demand, and above 30% needs intervention. However, MAPE should be assessed alongside forecast bias — a forecast with 15% MAPE but consistent 10% over-forecast bias is more damaging than 20% MAPE with zero bias.

Question 4

How to calculate forecast bias?

Accepted Answer

Forecast Bias = Average of (Actual - Forecast). Positive bias means chronic under-forecasting (stockout risk), negative bias means chronic over-forecasting (excess inventory). Bias should ideally be within plus or minus 5% of average demand. Track bias alongside the tracking signal (CFE/MAD) — when tracking signal exceeds plus or minus 4, the forecast is systematically biased.

Demand Forecasting Diagnostic Tool

Upload Sales Data Excel / CSV

Enter Sales Data Manually Up to 10 SKUs

Related Mathnal Resources