How do you choose the best demand forecasting method for MRP?

Manufacturers supplying pasts and components often encounter irregular demand patterns. Based on variances in demand volume and occurrences, intermittent demand types can be classified as smooth, lumpy, erratic and intermittent. Traditional forecasting algorithms perform poorly on intermittent demands. So the first step should be to check the degree of intermittence in your demand profile

O padr?o de demanda é a maneira como sua demanda flutua ao longo do tempo, como sazonal, cíclica, de tendência ou aleatória. Diferentes padr?es de demanda exigem diferentes métodos de previs?o, pois alguns s?o mais previsíveis e estáveis do que outros. Por exemplo, se sua demanda for sazonal, você pode usar um método de série temporal que capture as varia??es sazonais e as ajuste para elas. Se sua demanda for aleatória, você pode usar um método causal que identifique os fatores que influenciam sua demanda e quantifique seu impacto.

in order to choose the most accurate forecasting tool for the product you have to understand the demand pattern and they are seasonal, random, trend or combined patterns (Seasonal & Trend). understanding demand pattern is important because forecasting methods serve in different ways. for example moving average forecast could be good for trend and random variation but the issue is that it lags the changes in the pattern as it uses historical data. but if the pattern is well understood like why it's changing and when changes are a predicted like a cyclical pattern we can use weighted moving average for example to avoid lagging the changes for more accuracy.

Data availability refers to the amount and quality of historical and current data accessible for demand forecasting. It directly influences the accuracy and reliability of forecasts, as well as the complexity and cost of the forecasting method used. With abundant and high-quality data, statistical or machine learning methods can be applied to identify patterns and relationships, producing more accurate forecasts. However, with limited or unreliable data, qualitative or judgmental methods, such as expert opinions, surveys, or market research, may be necessary to inform demand projections.

forecast horizon can be short, medium or long-term but it has to be longer than the lead time. Because reorder point = DDLT (Demand during lead time) + SS (Safety Stock) - SOH (Stock On Hand). DDLT basically is the forecast if the LT is 4 weeks and the forecast horizon is 1 week in this case shortage or stock out is expected due to forecast error.

Forecast evaluation is important to improve the accuracy of the forecast. and there are many ways to measure the accuracy of the forecast. Absolute Deviation Quantity: Actuals - Forecast ? MAPE. measures the average of absolute errors or deviations percentages across specific period. Tracking Signal = Algebraic Sum of errors / MAPE MSE (Mean Squared Error) = It magnifies the error to discover top contributors Bias = occurs when there are consistent differences between actual and forecast of those quantities across planning period and when bias is zero that means random variations exist.