Exam Notes on Forecasting and Supply Chain PDF

# Push and pull boundry - Pull - initiated in response to a customer order - Reactive process - High production uncertainty - short production lead times - Service level and customer satisfaction -> product customization - Responsiveness - Flexible and responsive approaches...

# Push and pull boundry - Pull - initiated in response to a customer order - Reactive process - High production uncertainty - short production lead times - Service level and customer satisfaction -> product customization - Responsiveness - Flexible and responsive approaches - Push - initiated in anticipation of a customer order - speculative process - build your inventory - Low production Uncertainty - Long Production Lead Times - Cost Minimization - Resource allocation - products that are easy to forecast, e.g. water, electricity - more homogenous, necessity, e.g. water, electricity - Push/Pull boundry seperates push processes from pull processes ## Effect of more pull rather than push process? - Reducing customer demand uncertainty ## Conflicting Objectives in the Supply Chain | Purchasing | Manufacturing | Goals | Customers | | ----------------- | ------------------------- | ---------------- | ----------------- | | Stable volume requirements | Long run production | High quality | High productivity | | Flexible delivery time | High productivity | High productivity | Low production cost | | Little variation in mix | Low production cost | | Short order lead time | | Large quantities | | | High in stock | | | | | Enormous variety of products | | | | | Low prices | | Low inventory | | | | | Reduced transportation costs | | | | | Quick replenishment capability | | | | # Maximize SC surplus SC Surplus = Consumer Value - SC Cost = Consumer Value - Price -> Consumer Surplus + Price - SC Cost -> SC Profitability **NOTE that the consumer is the only source of revenue in a SC** # Forecasting ## The purpose of forecasting - is to explain as much of the systematic variability as possible and - describe or quantify the unsystematic variability to support a decision. ## Facts - A good forecast is a range not a number. - Long-term forecasts are less accurate - Aggregate forecasts are more accurate - forecast for more than one product, time period or location **Bullwhip effect** -> the effect of information sharing, information sharing decreases the total demand error ## Types of forecasting - qualitative - subjective methods based on intuition and experience - casual models - data based where the are be a cause and effect relation between variables - time series - based on historical data and assume the past indicates the future # Time series: the value of the same variable is recorded at regular intervals - **Level:** an 'average' around which observations vary - **Trend:** a predictable increase or decrease in the level over time - **Seasonality:** a pattern of predictable and recurring shifts in the level - **Random noises:** unpredictable variations in the data A time series with no systematic variability is also called stationary time series - For stationary time series, the forecast for all future periods is the same # Moving average - Bigger N - More responsive and variable (volatile) - Smaller N - More smooth # Exponential smoothing - Small alpha - more smooth - better for stable demand - demands with a lot of noise - Bigger alpha - more variable, chases demand (responsive) - better for changing demand # Forecasting error: ## Measures of accurance: - MAD: mean absolute deviation - average of absolute errors - Forecast is good if MASD is as soon as possible - MSE: mean squared error - Average of the forecast errors squared - Large errors are disproportionately more "expensive" than small errors - MAPE: Mean absolute percentage error - average of the absolute percentrage errors - The forecast accuracy is often defined as 1-MAPE **MAD₁ < MAD2** **МАРЕ₁ < МАРЕ2** **Or** **MAD₁ < MAD2** **MSE₁ < MSE2** ## Measures of bias: A forecast is biased when it consistently over (under) estimates the demand - MFE: Mean forecast error - RSFE: Running sum of forecast errors - Positive (negative) values of MFE or RSFE tell us that the forecast is too optimistic (pessimistic) If MFE or RSFE are “close to zero”, then the forecast is unbiased. - To quantify we use the tracking signal **TS =MFE/MAD** where a forecast is unbiased if **TS ∈ [-0.5,0.5]** Consider a forecast: If the mean absolute deviation (MAD) is close to zero, does it mean the Mean Forecast Error (MFE) is also close to zero? Answer: Yes! Explanation: Referring to the definitions of MAD and MFE, since for any error et, -|et| ≤ et ≤ |et|, we know that - MAD < MFE < MAD. ■ If MAD is close to zero, so is –MAD. So MFE is also close to zero. # Inventory management 1) Economies of scale* - Use everyday. Due to the fixed ordering/transportation cost, it is often cheaper to order in large batches 2) Uncertainties* (you cannot reduce or eliminate uncertainty) - Buffer against uncertainty. Uncertainties in demand, lead time, supply (production break-down, quality), etc. 3) Speculation - hedging against price increases (e.g., buying excess crude oil when the price is low) 4) Smoothing - To manage seasonal demand patterns and capacity limit. 5) Transportation - “Pipeline inventory" e.g., Ocean transport takes one month ## Why hold inventory? ## Inventory holding cost: the sum of all costs that are proportional to the amount of inventory physically on hand at any point of time - Maintanance and handling - warehouse costs - opportunity costs - obsolescence - insurance - investement in warehouse Which of the following is NOT a component of inventory holding cost? A. Cost of storage space B. Product obsolescence C. Loss of customer goodwill D. Opportunity cost of alternative investment Answer: C Explanation: Lost of goodwill is a cost incurred by not holding enough inventory and thus stock-out occurs. It happens when demand is uncertain. We will explain this in more detail in the newsvendor model. # Observations under Q* 1. The annual holding cost = annual ordering cost 2. Total ordering and holding costs are relatively stable around Q* -> we can order a convenient lot size close to Q* instead of the precise EOQ 3. Q* increases with demand D, ordering cost S and decreases with holding cost H 4. If demand D or ordering cost S changes to_k * D or k * S, then Q* changes to √k Q* 5. If the unit holding cost H changes to k * H, then Q* changes to √k Given everything the same, EPQ is bigger than EOQ because it has lower inventory cost # Newsvendor Model: - The simplest model that captures the essence of inventory management under demand uncertainty - Applicable to products with short lifecycle and uncertain demand ## Discrete demand: - In general, we want to find the maximum Q so the expected profit is bigger than or equal to the expected loss, i.e., **Pr(D≤Q) > Cu / (Cu + Co)** - Probability of not running out of stock - Cycle service level (CSL) ## Loss of goodwill - The cost of running out of stock may include the loss of customer goodwill. In accounting, goodwill is an intangible asset of companies ## Facts: - The optimal order quantity is not necessarily equal to average forecast demand - The optimal quantity depends on the relationship between marginal profit and marginal cost - As order quantity increases, average profit first increases and then decreases - As production quantity increases, risk increases. In other words, the probability of large gains and of large losses increases # Continuous Demand Newsvendor model with normal demand **Q* = μp + z* ση** - Expected Demand (SS) - Safety Stock Because D is continuous, we can always find a Q* such that marginal gain = marginal cost # Analysis of Q and z w.r.t. CSL If CSL* = 0.5, area of the curve for LHS=RHS=0.5 - z* = 0 - * Q* = μp (just order the avg. D.) - *Q* is independent of σ - *Q* is increasing in σ If desired CSL is higher than 0.5, CSL* > 0.5 - z* > 0 - *Q* > μD - *Q* is increasing in σ If desired CSL is lower than 0.5, CSL* < 0.5 - z* < 0 (safety factor is negative!) - *Q* < μD - *Q* is decreasing in σ # Demand forecast - Remember from forecasting: demand has a systematic component and an unsystematic (random) component - µD is the systematic component from a point forecast (see forecasting), e.g. moving average, simple exponential smoothing, Holt's method, - σD = 1.25 * Mean Absolute Deviation is the unsystematic (random) component (determine MAD out of historical data) ## Facts: - SS increases at an increasing rate with CSL - The higher the targeted service level, the higher the safety stock. - Decreasing MAD results in lower SS # Supply Chain Coordination Contract ## Supply Chain Coordination - each firm's objective becomes aligned with the supply chain's objective and optimal performance is achieved ## Revenue-Sharing Contract - The supplier sells products to the retailer at a lower wholesale price - The retailer returns a fraction *φ* of his revenue to the supplier - The contract works well in cases of demand like Blockbuster, such as clothing, digital music industries - A drawback of the Revenue Sharing Contract is the need for the Manufacturer to monitor Retailer's revenues. - Thus, it is important that the administrative burden to monitor data is not excessive (e.g., electronic exchange of POS data) ## Buy-back contract - It entails less risk - It is not suitable if the suppliers are too far away, the transportation cost is too high, or the product's salvage value is close to zero # Summary: - Supplier and retailer can achieve higher profits if they align their incentives and coordinate the Supply Chain - The rational of a coordinating contract is to distribute the risks (for instance, the inventory risk) along the Supply Chain Question: Suppose that the retailer faces customer demand with a continuous distribution, while all others remain the same. If the retailer enters a buy-back contract with the supplier. What is the buy-back price so that the supply chain can achieve the same order quantity as in an integrated supply chain (by rounding the result to the nearest integer)? ## Solution: When demand follows a continuous distribution, then one critical ratio corresponds to one order quantity. So, to make the order quantities the same under continuous distribution, **we only need to make the critical ratios under the two scenarios the same.** ## Revenue-sharing vs buy-back contract **Revenue-sharing** - Revenue-sharing works well in cases of peak demand such as clothing, digital music industries - A drawback of the Revenue Sharing Contract is the need for the Manufacturer to monitor Retailer's revenues. - Thus, it is important that the administrative burden to monitor data is not excessive (e.g., electronic exchange of POS data) **Buy-back contract** - It entails less risk - It is not suitable if the suppliers are too far away, the transportation cost is too high, or the product's salvage value is close to zero