Forecasting of Core Returns for Remanufacture: A Time Series Analysis

摘要

Over centuries, consumption of natural resources has been on a steady increase in re-sponse to the increasing global population. Increased and unsustainable use of natural re-sources in addition to increased manufacturing is a ecting the environment adversely. Hence, governments and environmental protection agencies are implementing rm regulations for industries to reduce their footprint on environmental pollution, for instance by ensuring that their waste products are not only disposed sustainably but also reduced. In response to these regulations, industries have embraced product end-of-life management strategies. These include reverse logistic, material and product recovery, reusing, recyling and remanufacturing.;This Thesis addresses one of the major challenges in remanufacturing which is uncertain-ties in the number of core returns for remanufacture. Speci cally, we propose a time series model that uses real data from a partner International OEM company that manufactures aswell as remanufactures electronic products. A unique aspect of the data that was obtained was the fact that speci c distinctions were made delineating billable return products from warranty return products for remanufacture. It is with this uniqueness that we sort to con-struct three time series model that is (a) Overall product core return; (b) Warranty return and (c) Billable return.;The forecast for the overall product core return and billable return was calculated using the Seasonal ARIMA (autoregressive integrated moving average) model, whereas the war-ranty return forecast was calulated using the ARIMA model. The best model was selected on the basis of akaike information criterion. ARIMA(0,1,1)(0,1,0) was selected as the best model for overall returns; ARIMA(0,1,1) was selected as the best model for warranty return and ARIMA(0,1,0)(0,1,0) was selected as the best model for billable return. The se-lected models were proven to be appropriate by means of residual diagnostics which includes Box-Ljung test, residuals of ACF, ARCH e ect and Jarque Bera test. Two-thirds of the data was used to build the models. After veri cation, this models were used to forecast the remaining one-third of the data. The accuracy of these forecasting results were determined with ME, RMSE, MAE, MPE, MAPE, MASE and ACF1. Overall, though not generizable to all companies, our model proved that for our partner company the overall returns were largely driven by the billable returns hence making it a pro table venture.