b. Do some assessment of the inventory implications of different lead times . Recall that if Shure follows a base stock policy targeting a 93 % fill rate , the amount of safety stock inventory they would need for an L-period lead time is approximately equal to the 95" percentile of a standard normal distribution times sqrt (L + 1) times the standard deviation of the forecast error (your RMSE estimate ). Put more directly: safety stock = 1.64 *sqit (L+1)*RMSE . [Deliverable: Fill in table in template . Add a few sentences to comment on the inventory impact of different lead times , and relate that to the accuracy numbers in part a.] (Postponement Analysis ) Another program being considered at Shure is to try to implement a kind of postponement strategy - delivering a semi -finished good to the US distribution center and then completing final configurations there (i.e., postponing the decision of which h exact product to produce until later in the supply chain ). Such a program would effectively mean that Shure can restrict attention to TOTAL demand for the whole product line (aggregated across all 9 products ), not demand for the individual products (since final configurations could be completed very quickly ). Therefore a. Compare the inventory requirements between using and not using a postponement strategy , assuming that in both cases Shure wants a 95 % fill rate and uses a 3- month forecast window . (Hint - you've done the necessary calculations in question 1 above.) [Deliverable : Using the information you gathered in question 1, write a few sentences commenting on the inventory impact of going with postponement .] II. Dish Network Data This problem uses the Dish Network data , which can be found on the course web site . Use R code to obtain a time series containing the total number of boxes (combining both return locations ) returned each week for all the weeks in the data set . (Use Tableau to extract the weekly data and export or paste it into Excel . Then save the result in CSV format with file name DishWeekly.csv so it can be read into R.) Given the way the data were collected, some weeks will not contain "representative " values - i.e., the first several weeks and last several weeks have low returns because only outgoing shipments from January through June are counted . Use the week beginning January 26 as the initial "representative week ", and the week beginning July 6 as your final week - discard the rest of the weeks . Split the remaining data into a "training " set consisting of the first 14 weeks and a "test " set consisting of the last 10 weeks . Then select an appropriate exponential smoothing model (simple exponential smoothing , smoothing with trend , Holt - Winters ) to use on the data . Use that model to generate forecasts for the data set (after dropping the "non -representative " weeks ), and report the RMSE that you get when applying your model (with one week ahead forecasts) to the "test" data. [Deliverable : Fill in table in template .]
