ISE 754 (Fall 2026) 6.2 Periodic Safety Stock
Overview
Michael G. Kay compares periodic base-stock and reorder-point policies, distinguishing single-unit replenishment without setup costs from batch ordering triggered at a minimum inventory level. The lecture explains lost-sales and backorder models, Julia-based event simulation, Poisson demand and the PASTA property, and bootstrapping; it then extends the discussion to multi-echelon supply chains and the Beer Game.
Key takeaways
- Base-stock policies choose Qmax to cover demand between replenishments and assume unit orders have no setup cost; reorder-point policies choose Qmin and trigger a batch order because each order incurs setup cost.
- The base-stock lost-sales objective accounts for carrying cost on average inventory Q-bar and the probability of stockout pi-zero, unlike a one-time safety-stock model that accounts for the cost of a lost sale.
- An event-calendar simulation in Julia can represent demand and replenishment arrivals in chronological order, making it flexible enough for inventory control logic that is difficult to solve analytically.
- When Poisson demand is appropriate, PASTA means Poisson arrivals see time averages, allowing the probability of stockout experienced by customers to be estimated from the system's time-based stockout probability.
- Bootstrapping samples observations with replacement, allowing limited demand data to generate repeated samples for estimating inventory statistics.
- In a multi-echelon chain, retailer stockouts can cause lost sales, while upstream wholesaler and distributor shortages are more commonly handled as backorders.
Chapters
- A base-stock policy selects a target quantity Qmax to cover demand between replenishments, assuming no setup cost and allowing orders of one unit.
- A reorder-point policy instead selects a minimum level Qmin and orders a batch when inventory reaches it, reflecting a setup cost per order.
- The base-stock control model processes demand and replenishment arrivals as events; it can represent lost sales, where an unavailable product means the customer leaves, or backorders, where the customer receives it later.
- The lost-sales objective includes inventory carrying cost based on average inventory Q-bar and the probability of being out of stock, pi-zero.
- A Julia simulation sorts customer demands and replenishment arrivals in an event calendar, then processes events in time order to model inventory behavior.
- When demand data are unavailable and a Poisson assumption is appropriate, an estimated average arrival rate can support analysis without the usual simulation data.
- The Poisson property called PASTA, Poisson arrivals see time averages, lets customer arrivals observe the same stockout probability as a time-based view of the system.
- Bootstrapping creates repeated demand samples by random sampling with replacement, helping characterize statistics from limited observations; the bakery example estimates carrying cost from bread losing 10% of its value per day.
- Reorder-point control tracks inventory position, including both stock on hand and units already on order, and supports lost-sales or backorder variants.
- In a multi-echelon chain, a retailer may order from a wholesaler, which orders from a distributor and ultimately a factory.
- Lost sales are most relevant at the retailer, where a customer can leave without buying; upstream wholesalers and distributors typically backorder retailer orders instead.
- The lecture closes with an introduction to the Beer Game, using a four-echelon supply chain to prepare students for the in-class exercise.
Transcript excerpt
0:02 In this lecture, we'd be looking at periodic safety stock. This table provides an overview of both the previous lecture and then this lecture. In the previous lecture, we were looking at the case of a one-time safety stock where we had the decision variable Q. For this lecture, we'll be looking at two different models. The first is base stock, which is a pure safety stock model where we determine a decision variable Q max. We're basically trying to determine how much inventory do we want to keep on hand to cover the demand that can occur between replenishments. And then the second model is the reorder point. With the reorder point, we want to determine
0:50 how low does our inventory need to get, which is Q min. And then when it gets to that level, we actually order more than just a single unit. We order a batch of product. The reason for that is there's some associated setup cost with every time we order. And the assumption then is for base stock that there is no setup cost. We can order one unit for the same cost as ordering multiple units. This figure shows the basic control logic for base stock. Basically, two events happen. We either have a demand that occurs or we have a replacement that arrives, something that was ordered uh previously. And then whatever the lead
1:38 time is, once it arrives, that's another event. This is the basic model for base stock and the scenario covered in this model is the case of lost sales. The other model, model two, handles the case of backordering. The difference is when a customer arrives and no product. With lost sales, we just lose the sale. With backorder, the customer still gets the product, but it's backordered and the customer
This is the start of the transcript. Save this video free for the whole transcript, searchable and time-coded.
In this course
Related notes
Summary, takeaways, and chapters were generated by AI from the video's transcript and may contain errors. The video belongs to its creator, Michael G. Kay.