Sponsor:

Invistics Corporation

Research Team:

Manuel D. Rossetti

Universities Involved:

University of Arkansas

Start Date:

08/01/09

End Date:

07/31/10

Summary:

The main goal of this research project is to investigate and develop methods that allow target service levels (e.g. fill rate, ready rate) to be met for planned inventory policies and lot-sizing rules using retrospective or bootstrapping based simulation procedures. Safety stock levels can be calibrated via simulation procedures to increase the likelihood that the planned for performance will actually be met in practice. Gudum and de Kok (2002) proposed a simulation based safety stock adjustment procedure that is motivated by the problem of comparing lot-sizing rules. We refine the procedure by presenting a better algorithmic representation so that practitioners can more readily understand and implement the procedure. We extend the study by considering the problem of comparing other lot-sizing rules; namely, lot-for-lot, part-period balancing and least unit cost. Experiments illustrate how the safety stock adjustment procedure performs under a number of demand scenarios.
Inventory policies and lot-sizing rules determine when to order and how much to order over a planning horizon. For example assume that an item is being controlled by a periodic (r, q) inventory policy. A standard procedure for applying this model is to use historical data to forecast the mean demand and a measure of uncertainty concerning the forecast (e.g. mean squared error of the forecast). These values are then used along with assumptions concerning the lead-time to estimate the moments of an assumed family for a lead-time demand distribution. This lead-time demand distribution is used to set the policy parameters to be used over the planning period to meet a particular service level. As outlined in Wagner (2002), there are inherent representational errors and conceptual problems with this approach. One of the key problems is that this process is unable to adequately represent the true demand characteristics, especially in the tail. Another problem with this approach is that it assumes a stationary demand pattern. Wagner (2002) also points out that the data is often “dirty”, violating many assumptions associated with the forecasting technique, the model fitting procedures, and the inventory model. When the policy is put in place and actual demand occurs, the planned for performance levels are not met.

Now, suppose we have a known sequence of demand (possibly from historical data, possibly a forecast, or possibly a sequence of future requirements as determined by a MRP lot-sizing procedure). Then, given an initial starting inventory level and a lead-time, the performance of the policy can be recreated or simulated over time. From the sample path, the operational service level (e.g. fill rate, ready rate, etc.) can be computed. Wagner (2002) argues that a retrospective simulation using real data can be used to calibrate the inventory control policy or safety stock levels to increase the likelihood that the planned for performance will actually be met in practice.

Thus, the main goal of this proposed research project is to investigate and develop methods that allow target service levels (e.g. fill rate, ready rate) to be met for planned inventory policies and lot-sizing rules using retrospective or bootstrapping based simulation procedures.