Why would you want to predict your production volumes?  If you can build a model to forecast the number of units of a certain product that you will produce, then you can:

• more accurately predict the amount of raw materials and component parts needed
• reduce your inventory
• better plan the labour resources needed on the factory floor
• manage your overtime
• provide more accurate forecasts to your suppliers

Background: My client produces three value streams (product lines) on the factory floor. Let's call them VS1, VS2 and VS3. In the past, they have only been forecasting the total production of their value streams, i.e., the total of VS1, VS2 and VS3. They are now interested in discovering what forces drive the demand for each of these value streams.

Analysis:  Below is a plot of the number of items produced on the factory floor each month for VS1, VS2 and VS3. Looking at VS1, we can see that we have two data points that are questionable (shown in circles on the pink line below). After investigating the source data for these points, they were found to be incomplete data and were omitted from the data set.

It is thought that the demand for these value streams is related to their overall market. The overall market can be divided into 5 groups. Two different sets of marketing data were obtained which show the industry production volumes for Groups 1 through 5 each quarter. In performing a regression analysis of each value stream with each industry group, we can examine the statistical output to determine which sets are statistically significant. In the case of this client, VS1 showed a strong relationship with Group 5 and VS2 and VS3 showed no statistically significant relationship to Groups 1 through 5. Details of the model measurements like R-squared and the Mean Absolute Deviation are omitted since this newsletter is meant more as a demonstration of modeling capabilities than a full mathematical paper.

What does this mean in plain English? The company now knows that the demand for VS1 is closely related to the number of Group 5 products purchased in the market. Based on market forecasts, they can now predict the demand for VS1. Looking at the graph below, the blue line is the actual production and the red and green lines are two different models which were developed. The vertical black line marks today's date and anything to the right of the line is a future forecast.

What about VS1 and VS2? While it is thought that at least a weak relationship to the market groups must exist, the relationship is not strong enough to create a regression model. Since we have 10 years of production history, we can attempt a Time Series analysis instead.