Thomas R. Willemain

A new approach to forecasting intermittent demand for service parts inventories

Abstract A fundamental aspect of supply chain management is accurate demand forecasting. We address the problem of forecasting intermittent (or irregular) demand, i.e. random demand with a large proportion of zero values. This pattern is characteristic of demand for service parts inventories and capital goods and is difficult to predict. We forecast the cumulative distribution of demand over a fixed lead time using a new type of time series bootstrap. To assess accuracy in forecasting an entire distribution, we adapt the probability integral transformation to intermittent demand. Using nine large industrial datasets, we show that the bootstrapping method produces more accurate forecasts of the distribution of demand over a fixed lead time than do exponential smoothing and Croston’s method.

Forecasting intermittent demand in manufacturing: a comparative evaluation of Croston's method

Abstract Intermittent demand appears at random, with many time periods having no demand. Manufacturers perceive the forecasting of intermittent data to be an important problem. In practice, the standard method of forecasting intermittent demand is single exponential smoothing, although some production management texts suggest the lesser-known alternative of Crostons method [Croston J.D., 1972, Forecasting and stock control for intermittent demands, Operational Research Quarterly , 23(3), 289-303]. We compared the two methods, using artificial data created to violate Crostons assumptions and real-world data from industrial sources. We conclude that Crostons method is robustly superior to exponential smoothing and could provide tangible benefits to manufacturers forecasting intermittent demand.

Model-based and model-free control of autocorrelated processes

Advances in automated sampling technology have made autocorrelated data commonplace. Positive autocorrelation degrades control charts designed by classical methods. If a correct time-series model of the autocorrelated process is available, many have adv..

Average run lengths for CUSUM control charts applied to residuals

A common approach to building control charts for autocorrelated data is to apply classical SPC to the residuals from a time series model of the process. However, Shewhart charts and even CUSUM charts are less sensitive to small shifts in the process mean when applied to residuals than when applied to independent data. Using an approximate analytical model, we show that the average run length of a CUSUM chart for residuals can be reduced substantially by modifying traditional chart design guidelines to account for the degree of autocorrelation in the data.

DESIGNING CONTROL CHARTS USING AN EMPIRICAL REFERENCE DISTRIBUTION

We address a problem of increasing practical interest: control charting in the absence of an assumed normal distribution. We analyze the sampling uncertainties inherent in selecting control limits from the empirical distribution of a large sample of wha..

Visualization and the process of modeling: a cognitive-theoretic view

are artifacts used to better understand our world. In the context of mining data to create knowledge, the modeler is often faced with discovering and understanding relationships in data that have no apparent analog in the laws of physical science. Much of these data are increasingly abstract; not directly related to the physical, perceivable world. The use of sketches and diagrams as aids in problem solving and as a means of communication is as old as recorded history. Can visualization help us not only to discover the patterns and relationships in these data, but also to use newly discovered knowledge to build computational models? The purpose of this paper is to provide a description of the process of modeling and, based upon theories of cognition, show how visualization can assist in developing and assessing computational models for very large, high dimensional data sets.

Graphical adjustment of statistical forecasts

Abstract Twelve subjects made judgmental adjustments to graphs of statistical forecasts. The data were artificial ARIMA series with known optimal forecasts. When statistical forecasts were nearly optimal, adjustment had little effect. When the forecasts were less accurate, adjustment improved accuracy. These results suggest the value of graphical review and adjustment of statistical forecasts.

Alumni Giving: The Influences of Reunion, Class, and Year.

We analyzed fifty years of inflation-adjusted data on the Annual Giving program of Princeton University. Most of the variation in both average size of gifts and percentage of class giving can be explained with simple models having three factors: reunion number, class identity, and fiscal year. Besides providing insights into factors influencing donations, these models provide a way to unmask features that are not evident in the raw data, such as trends in giving behavior and exceptional performances by particular classes in particular years.

The Economics of Safety Deregulation: Lives and Dollars Lost due to Repeal of Motorcycle Helmet Laws

Between 1976 and 1980, 28 state legislatures in the United States repealed or weakened their motorcycle helmet-use laws. This paper estimates the number of excess deaths attributable to this deregulatory activity, and the associated economic costs to society. Because of data limitations, no attempt was made to estimate the excess nonfatal injuries and associated costs. We applied a variant of log-linear contingency-table analysis to the monthly counts of motorcycle fatalities in the 48 contiguous states over the period 1975 through 1980. This analysis produced estimates of the total number of deaths, in each of 36 age-sex groups, that could be attributed to changes in the helmet laws. We then estimated the direct and indirect economic costs associated with fatalities in each age-sex group. Our findings indicate that 516 excess deaths occurred in 1980 in the 28 states that weakened or repealed their helmet laws. This represented 24 percent of the total motorcycle fatalities occurring in those states. Women and younger cyclists of both sexes comprised a disproportionate share of excess deaths. The economic costs to society that are associated with the excess fatalities resulting from the repeals of helmet laws total at least

The threshold bootstrap and threshold jackknife

180 million.