当前位置:首页 >> >>

The Top-Dog Index A New Measurement for the Demand Consistency of the Size Distribution in


THE TOP-DOG INDEX: A NEW MEASUREMENT FOR THE DEMAND CONSISTENCY OF THE SIZE DISTRIBUTION IN PRE-PACK ORDERS FOR A FASHION DISCOUNTER WITH MANY SMALL BRANCHES
¨ ¨ ¨ SASCHA KURZ, JORG RAMB

AU, JORG SCHLUCHTERMANN, AND RAINER WOLF

arXiv:0804.1412v1 [math.OC] 9 Apr 2008

Abstract. We propose the new Top-Dog-Index, a measure for the branch-dependent historic deviation of the supply data of apparel sizes from the sales data of a fashion discounter. A common approach is to estimate demand for sizes directly from the sales data. This approach may yield information for the demand for sizes if aggregated over all branches and products. However, as we will show in a real-world business case, this direct approach is in general not capable to provide information about each branch’s individual demand for sizes: the supply per branch is so small that either the number of sales is statistically too small for a good estimate (early measurement) or there will be too much unsatis?ed demand neglected in the sales data (late measurement). Moreover, in our real-world data we could not verify any of the demand distribution assumptions suggested in the literature. Our approach cannot estimate the demand for sizes directly. It can, however, individually measure for each branch the scarcest and the amplest sizes, aggregated over all products. This measurement can iteratively be used to adapt the size distributions in the pre-pack orders for the future. A real-world blind study shows the potential of this distribution free heuristic optimization approach: The gross yield measured in percent of gross value was almost one percentage point higher in the test-group branches than in the control-group branches.

1. Introduction The ?nancial performance of a fashion discounter depends very much on its ability to predict the customers’ demand for individual products. More speci?cally: trade exactly what you can sell to your customers. This task has two aspects: o?er what your customers would like to wear because the product as such is attractive to them and o?er what your customers can wear because it has the right size. In this paper, we deal with the second aspect only: meet the demand for sizes as accurately as possible. The ?rst aspect, demand for products, is a very delicate issue: Products in a fashion discounter are never replenished because of lead times of around three months. Therefore, there will never be historic sales data of an item at the time when the order has to be submitted (except for the very few “never-out-of-stock”-items, NOS items, for short). When one considers the knowledge and experience of the professional buyers employed at a fashion discounter—acquired by visiting expositions, reading trade journals, and the like—it seems hard to imagine that a forecast for the demand for a product could be implemented in an automated decision support system at all. We seriously doubt that the success of fashion product can be
2000 Mathematics Subject Classi?cation. Primary: 90B05; Secondary: 90B90. Key words and phrases. revenue management, size optimization, demand forecasting, Top-Dog-Index, ?eld study, parallel blind testing. This research was supported by the Bayerische Forschungsstiftung, Project DISPO—A Decision Support for the Integrated Size and Price Optimization.
1

2

¨ ¨ ¨ SASCHA KURZ, JORG RAMBAU, JORG SCHLUCHTERMANN, AND RAINER WOLF

assessed by looking at historic sales data only. In contrast to this, the demand for sizes may stay reasonably stable over time to extract useful information from historic sales data. In the historic sales data the in?uences of demand for products and demand for sizes obviously interfere. Moreover, it was observed at our partners’ branches that the demand for sizes seems not to be constant over all around 1 200 branches. The main question of this work is: how can we forecast the demand for sizes individually for each branch or for a class of branches? 1.1. Related Work. Interestingly enough, we have not found much work that exactly deals with our task. It seems that, at ?rst glance, the problem of determining the size distribution in delivery pre-packs can be considered as simple regression once you have historic sales data: Just estimate the historic size pro?le and ?t your delivery to that. At least two trivial US-patents [6, 8] have recently been granted and published along these lines (which witnesses that the US patent system may not have employed the necessary expertise in their patent evaluations . . . ). In our problem, however, the historic sales data is not necessarily equal to the historic demand data, and it is interesting how to ?nd the demand data in the sales data in the presence of unsatis?ed demand and very small delivery volumes per branch and per product. The type of research closest to ours seems to be classi?ed as assortment optimization. In a sense, we want to decide on the start inventory level of sizes in a pre-pack for an individual product in an individual branch. (Let us ignore for a moment that these unaggregated inventory levels are very small compared to other inventory levels, e.g., for grocery items.) Mostly, the successful approaches deal rather with NOS items than with perishable and not replenishable fashion goods. For example, assortment optimization in the grocery sector [4, Section 4] — one of the very few papers documenting a ?eld study — can usually neglect the e?ects of stockout substitution in sales data, which make demand estimation from sales data much more reliable. There is work on the speci?c in?uence of substitution on the optimization of expected pro?t (see, e.g., [7]), but the problem of how to estimate demand parameters from low-volume sales data in the presence of stock-out substitution remains. Much more work has been published in the ?eld of dynamic pricing, where in one line of research pricing and inventory decisions are linked. See [3, 2] for surveys. A common aspect of all cited papers (and papers cited there) that separates their research from ours is the following: those papers, in some sense, postulate the possibility to estimate a product’s demand in an individual branch directly from sales data, in particular from sales rates. In our real-world application we have no replenishment, small delivery volumes per branch, lost sales with unknown or even no substitution, and sales rates depending much more on the success of the individual product at the time it was o?ered than on the size. Therefore, estimating future absolute demand data from historic sales data directly seems to need extra ideas, except maybe for the data aggregated over many branches. Size optimization can be found in a few o?ers of retail optimization systems, like 7thOnline (http://www.7thonline.com). It is not clear, on what kind of research these products are based and under which assumptions they work well. Our partner ?rm has checked several o?ers in the past and did not ?nd any optimization tools that met their needs. 1.2. Our contribution. The main result of this work is: a useful forecast for the demand for sizes on the level of individual branches is too much to be asked for, but historic information about which sizes have been the scarcest and which sizes have been the amplest ones can be obtained by measuring the new Top-Dog-Index (TDI). The TDI can be utilized in a dynamic heuristic

THE TOP-DOG INDEX

3

optimization procedure, that adjusts the size distributions in the branches’ corresponding pre-packs accordingly until the di?erence between the scarcest and the amplest sizes can not be improved anymore. The main bene?t of the TDI: it measures the consistency of historic supply with sizes with the historic demand for sizes in a way that is not in?uenced by the attractivity of the product itself. This way, we can aggregate data over all products of a product group, thereby curing the problem with small delivery volumes per branch and product and size. The potential of our TDI-approach is shown in a blind-study with 20 branches and one product group (womens’ outer garments). Ten branches randomly chosen from the 20 branches (testbranches) received size pre-packs according to our heuristics’ recommendations, ten received unchanged supply (control-branches). The result: One percentage point increase of gross yield per merchandise value for the test-branches against the control-branches. A conservative extrapolation of this result for our partner would already mean a signi?cant increase of gross yield. We have not seen any ?eld study of this type documented in the literature so far. The only documented yield management studies in the apparel retail business (e.g., for dynamic pricing policies) try to prove the success of their methods by showing that they would have gained something on the set of data that was used to estimate the parameters of the model [1]. Such tests are very far from reality: a by-product of our study is that in our case it makes absolutely no sense to compare gross yield data of a fashion discounter across seasons, because the di?erences between the yields in di?erent seasons are much larger than the di?erences caused by anything we are interested in. This was the reason for us to use the parallel blind testing instead. Further tests are planned by our partner on a larger scale in the near future. 1.3. Outline of the paper. In Section 2, we brie?y restate the real-world problem we are concerned with. In Section 3, we show our experience with straight-forward estimators for the demand distribution on sizes. In Section 4, we introduce the new Top Dog Index, which is utilized in Section 5 in a heuristic optimization procedure. Section 6 is the documentation of a ?eld study containing a blind testing procedure among two groups of branches: one supplied with and one supplied without the suggestions from the ?rst step of the optimization heuristics. We summarize the ?ndings in Section 7, including some ideas for further research. 2. The real-world problem In this section, we state the problem we are concerned with. Before that we brie?y provide the context in which our problem is embedded. 2.1. The supply chain of a fashion discounter. As in most other industries the overall philosophy of supply chain management in fashion retailing is to coordinate the material ?ow according to the market demand. The customer has to become the “conductor” of the “orchestra” of supply chain members. Forecasting the future demand is, therefore, crucial for all logistics activities. Special problems occur in cases like ours, when the majority of inventory items is not replenished, because the relationship between lead times and fashion cycles makes replenishments simply impossible. The resulting “textile pipeline” has strong interdependencies between marketing, procurement, and logistics. The business model of our real world problem bases on a strict cost leadership strategy with sourcing in low cost countries, either East Asia or South East Europe. The transportation time is between one and six weeks, economies of scale are achieved via large orders.

4

¨ ¨ ¨ SASCHA KURZ, JORG RAMBAU, JORG SCHLUCHTERMANN, AND RAINER WOLF

2.2. Internal stock turnover of pre-packs. The material ?ow in our problem is determined by a central procurement for around 1 200 branches. All items are delivered from the suppliers to a central distribution center, where a so called “slow cross docking” is used to distribute the items to the branches. Some key ?gures may give an impression of the situation: 32 000 square meters, 80 workers, 30 000 tons of garment in 10 million lots per year. Each branch is delivered once a week with the help of a ?xed routing system. This leads to a sound compromise between inventory costs and costs of stock turnover. There are two extreme alternatives for the process of picking the items. The retailer can either work with one basic lot and deliver this lot or integer multiples of it for every article to the branches, or he develops individual lots for the shops. At the beginning of our analysis most of the articles were picked in one basic lot, but more than 40 other constellations were used additionally. The costs for picking the items are not relevant for the following analysis because only minor changes of internal processes are necessary. Only in later phases of our project a detailed analysis of the picking costs will be needed. 2.3. The problem under consideration. Recall that the stock turn-over is accelerated by ordering pre-packs of every product, i.e., a package containing a speci?ed number of items of each size. We call the corresponding vector with a non-negative integral entry for every size a lot-type. In this environment, we focus on the following problem: Given historic delivery data (in terms of pre-packed lots of some lot-type for each branch) and sales data for a group of products for each branch, determine for each branch a new lot-type with the same number of items that meets the relative demand for sizes more accurately. In particular, ?nd out from historic sales data some information about the relative demands for sizes. We stress the fact that stockout substitution in the data can not be neglected since replenishment does not take place (unsatis?ed demand is lost and does not produce any sales data). We also stress the fact that we are not trying to improve the number of items delivered to each branch but the distribution of sizes for each branch only. 3. Some real-data analysis evaluating seemingly obvious approaches To deal with the problem described in Section 2 our partner has provided us some historic sales data for approximately 1 200 branches over a time period of 12 months. The task was to forecast the size distribution of the future demand for each branch. A possible size distribution is depicted in Figure 1. The most obvious way to determine a size distribution as in Figures 1 is to count the number of sold items per size and divide by the total number of sold items. Here we have some freedom to choose the day of the sale where we measure the amounts. We have to balance two competing facets. An early measurement may provide sales ?gures which are statistically too small for a good estimate while a late measurement may su?er from unsatis?ed demand that is not present in the sales data. The business strategy of our partner implies to cut prices until all items are sold. So, a very late measurement would only estimate the supply instead of the demand. As there is no expert knowledge to decide which is the optimal day of sales to count the amounts and estimate the size distribution we have applied a statistical test to measure the signi?cance of the size distributions obtained for each possible day of counting the sold items. Given a data set D, a day of sales d, and a size s let φs,d (D) be the estimated demand for size s measured on day d as described above. We normalize so that φs,d (D) = 1 for each day and
s

THE TOP-DOG INDEX

5

Table 1. A possible size distribution of the demand

each data set. Our statistical test partitions the original data set D randomly into two disjoint data sets D1 and D2 . Naturally we would not trust a forecast φ whenever φs,d (D1 ) and φs,d (D2 ) are too far apart. Statistically speaking φ would not be robust in that case. To measure more precisely how far apart φs,d (D1 ) and φs,d (D2 ) are, we de?ne the discrepancy δd as δd (D1 , D2 ) := |φs,d (D1 ) ? φs,d (D2 )|. In Figure 2 we have depicted the average discrepancy over all branches
s

for the ?rst 60 days of sale. For our practical application the signi?cance of φ is too low for all possible days of measurement. We remark that in our examined data set the discrepancy δd tends to 0.19 as d tends to the end of the sales period. An explanation why this obvious approach does not work in our case is due to the small sales numbers and the interference of product attractivity and price cutting strategies. Even when we tried to use only use ordinal information generated from an estimated size distribution (some size is too scarce, some size too ample), we encountered di?erent results from di?erent samples (same size was scarce in one sample and ample in the other). In the same manner we have checked some common parametric models (mostly based on an estimation of a constant sales rate for each individual product-size pair) for demand forecasting on historic sales data from literature. None of them gives signi?cant information of the size distribution of the demand of our data set, since the sales rates vary drastically and depend more on the attractivity of the products than on the sizes. For the details we refer to two diploma theses [5, 9] from our group. 4. The Top-Dog-Index (TDI) In the previous section we learned that in our application, ?rst, we cannot trust the common parametric models for the demand distribution and, second, measuring a size pro?le directly from sales data may lead to more or less random decisions. The main reason for the latter is the former and the interference between attractiveness of o?ered products and compatibility of o?ered sizes. Because of this, the stockout saturation of sales data happens at almost all times during the sales period, and thus aggregating the sales data of di?erent products yields no reasonable information. Our new idea throws overboard the desire to estimate an absolute size pro?le of the demand in every branch. Instead, we try to de?ne a measure for the scarcity of sizes during the sales process that can be estimated from historic sales data in a stable way.

6

¨ ¨ ¨ SASCHA KURZ, JORG RAMBAU, JORG SCHLUCHTERMANN, AND RAINER WOLF

Table 2. Average discrepancy δd over all branches

The following thought experiment is the motivation for our distribution free measure: Consider a product, for which in a branch all sizes are sold out at the very same day. This can be regarded as the result of an ideal balance between sizes in the supply. Our measure tries to quantify the deviation of this ideal situation in historic sales data. How can this be done? In the following, we extract data of a new type from the sales process. Fix a delivery period ? := [0, T ] from some day in the past normed to day 0 to day T . Let B be the set of all branches that are operating in time interval ?, and let P be the set of all products in a group delivered in time interval ? in sizes from a size set S. We assume that in each branch the product group can be expected to have homogeneous demand for sizes throughout the time period. Fix b ∈ B. For each p ∈ P and for each s ∈ S let θb (p, s) be the stockout-day of size s of product p, i.e., the day when the last item of p in size s was sold out in branch b. Fix a size s ∈ S. Our idea is now to compare for how many products p size s has the earliest stockout-day θ(p, s) and for how many products p size s has latest stockout-day θ(p, s). These numbers have the following interpretation: If for many more products the stockout-day of the given size was ?rst among all sizes, then the size was scarce. If for many more products the stockout-day of the given size was last among all sizes, then the size was ample.

THE TOP-DOG INDEX

7

In order to quantify this, we use the following approach. (In fact, it is not too important how we exactly quantify our idea, since we will never use the absolute quantities for decision making; we will only use the quantities relative to each other.) De?nition 4.1 (Top-Dog-Index). Let s ∈ S be a size and b be a branch. The Top-Dog-Count Wb (s) for s in b is de?ned as (1) Wb (s) := { p ∈ P | θb (p, s) = min θb (p, s ) }
s ∈S

and the Flop-Dog-Count Lb (s) in b is de?ned as (2) Lb (s) := { p ∈ P | θb (p, s) = max θb (p, s ) } .
s ∈S

Moreover, for a ?xed dampening parameter C > 0 let (3) TDIb (s) := Wb (s) + C Lb (s) + C

be the Top-Dog-Index (TDI) of Size s in Branch b. In the data of this work, we used C = 15. 4.1. Statistical signi?cance of the Top-Dog-Index. In a similar way as in Section 3 we want to analyze the signi?cance of the proposed Top-Dog-Index. Since this method is supposed to be applied to a real business case, we analyze the statistical signi?cance in more detail. Instead of two data sets D1 and D2 as in Section 3 we utilize seven such sets Di . Therefore, we assign a random number in {1, 2, 3, 4} to each di?erent product. The sets are composed of the data of products where the corresponding random number lies in a characteristic subset of {1, 2, 3, 4}. See Table 3 for the assignment. For the interpretation we remark that the pairs (D1 , D2 ), (D3 , D4 ), and (D5 , D6 ) are complementary. The whole data set is denoted by D7 . D1 {1, 2} D2 {3, 4} D3 {1, 3} D4 {2, 4} D5 {3} D6 {1, 2, 4} D7 {1, 2, 3, 4} Table 3. Assignment of test sets

Since the Top-Dog-Index is designed to provide mainly ordinal information, we have to use another statistical test to make sure that it yields some signi?cant information. Let TDIb (s, Di ) denote the Top-Dog-Index in Branch b of Size s computed from the data in Data Set Di . We ?nd it convincing to regard the ordinal information generated by the Top-Dog-Index as robust whenever we have (4) TDIb (s, Di ) TDIb (s , Di ) ?? TDIb (s, Dj ) TDIb (s , Dj )

for each pair of sizes s, s and each pair of data sets Di , Dj . In words: the order of Top-Dog-Indices of various sizes does not change signi?cantly when computed from a di?erent sample. The following

8

¨ ¨ ¨ SASCHA KURZ, JORG RAMBAU, JORG SCHLUCHTERMANN, AND RAINER WOLF

is a su?cient condition for this to happen: (5) TDIb (s, Di ) ≈ const i j TDIb (s, Dj )

Our ?rst aim is to provide evidence that the TDIb (s) values are robust measurements in this sense. There is a nice way to look at equation 5 graphically. Let us ?x a size s. For each branch b let us plot a column of the relative values PTDIb (s,Di )j ) for all branch-size combinations and for j TDIb (s,D all i. The columns corresponding to the same branch-size combination but di?erent samples are stacked on top of each other. This way, each colum stack has height one in total, and the size relations of the measurements based on the di?erent samples can be assessed right away. To show the robustness of the mean and the median, we have added an additional column for each of them: The median of the seven estimates corresponds to the height of the top-most column, the mean to the height of the second-to-top-most column. The goal now is not to read o? the exact values, but to get an intuitive impression how heavy an estimate depends on the data subset it was computed from. The Top-Dog-Indices are plotted this way in Figure 1.

Figure 1. Relative distribution of the Top-Dog-Index on seven data subsets for all branch-size combinations; the two top-most columns are median and mean, resp. In order to provide some intuitively clear reference data to compare to the plot of Figure 1, we present the corresponding plots for the two extreme cases of deterministic numbers (i.e., TDIb (s, Di ) = TDIb (s, Dj ) for all i and j) and totally random numbers in Figure 2. In the complete deterministic case the areas of same color form perfect rectangles. In the random case the areas of same color corresponding to the data subsets form zig-zag lines; the median has fewer zig-zag than the mean, but both are quite stable because the random numbers are all from the same distribution. It is immediately obvious that the plot of Figure 1 looks more like the plot in the deterministic case as the plot in the random case. It is interesting to note that the dampening parameter in the computation of the TDI does indeed in?uence the amount of noise in the plot but had almost no in?uence on the order of TDI values, which is what we indend to use.

THE TOP-DOG INDEX

9

Figure 2. Relative distribution of deterministic and random numbers on seven data subsets and branches; the two top-most columns are median and mean, resp.

Figure 3. Relative distribution of sales up to Day 0 on on seven data subsets for all branch-size combinations; the two top-most columns are median and mean, resp.

Figure 4. Relative distribution of sales up to Day 12 on on seven data subsets for all branch-size combinations; the two top-most columns are median and mean, resp.

10

¨ ¨ ¨ SASCHA KURZ, JORG RAMBAU, JORG SCHLUCHTERMANN, AND RAINER WOLF

Recall that we have analyzed the direct estimation of size distributions for each branch by looking at two complementary samples of the data set. In order to be able to directly compare the results of that attempt to Figure 1, we plot the size distributions for the same seven data subsets. This is shown in Figure 3 for an estimate from the sales up to Day 0 (the ?rst day in the sales period) and in Figure 4 for an estimate from the sales up to Day 12. It can be seen that Day 0-estimates are extremely dependent on the data subset even if only the relative values are taken into account, i.e., the estimates are not robust even in the weak sense measured in the plot. The Day-12 estimates are more robust, but not even close to what the TDI achieves in Figure 1. Moreover, as we said, those estimates are already quite close to the supply because of unsatis?ed demand, and will therefore fail to measure the size distribution of the demand. Generating a robust statistics for a set of data is, of course, easy: just assign the same deterministic number to each object. This is not what we want since this does not carry any useful information. We claim that the Top-Dog-Index exhibits di?erences between demands for sizes in each branch individually. That this is indeed the case, follows from the Top-Dog-Indices that we encountered in the ?eld study documented below. But let us ?rst discuss which actions we could take to improve the size distributions of the branches’ supply. So far we have argued that the Top-Dog-Index produces size-related information in a robust way, while other methods fail (at least for our given real-world data). In the next section we describe a procedure to harmonize demand and supply with respect to the size distribution. In Section 6 we provide evidence via a real-world blind study that this procedure helps to raise the gross yield in reality; thus, the Top-Dog-Index is correlated to the size distribution of the demand. 5. The heuristic size optimization procedure based on the TDI Interesting for us is not the absolute TDI of a size but the TDIs of all sizes in a branch compared to each other, i.e., the ordinal information implied by the TDIs. The size with the maximal TDI among all sizes can be interpreted as the scarcest (the one that was sold the fastest) size in that branch; the size with the minimal TDI among all sizes can be interpreted as the amplest size in that branch. Of course, we have the problem of deciding whether or not a maximal TDI is signi?cantly larger than the others. Since the absolute values of the TDIs have no real meaning we did not even try to assess this issue in a statistically profound way. Our point of view is again that absolute forecasting is too much to be asked for. Therefore, we resort to a dynamic heuristic optimization procedure: sizes with “signi?cantly” large TDIs (Top-Dog-Sizes) should receive larger volumes in future deliveries until their TDIs do not improve anymore, while sizes with “signi?cantly” small TDIs (Flop-Dog-Sizes) need smaller supplies in the future. Whenever this leads to oversteering, the next TDI analysis will show this, and we go back one step. This is based on the assumption that the demand for sizes does not change too quickly over time. If it does then optimization methods based on historic sales data are useless anyway. Let us describe our size distribution optimization approach in more detail. We divide time into delivery periods (e.g., one quarter of a year). We assume that the sales period of any product in a delivery period ends at the end of the next delivery period (e.g., half a year after the beginning of the delivery period). Recall that a size distribution in the supply of a branch is given by a pre-pack con?guration: a package that contains for each size a certain number of pieces of a product (compare Section 2.2). We want to base our delivery decisions for an up-coming period on ? the pre-pack con?guration of the previous period and ? the TDI information of the previous period giving us the deviation from the ideal balance

THE TOP-DOG INDEX

11

According to given restrictions from the distribution system, we assume that only one pre-pack con?guration per branch is allowed. We may use distinct pre-pack con?gurations for di?erent branches, though. Since we are only dealing with the size distribution of the total supply but not with the total supply for a branch itself, the total number of pieces in a pre-pack has to stay constant. Since the TDI information only yields aggregated information over all products in the product group, all products of this group will receive identical pre-pack con?gurations in the next period, as desired. In order to adjust the supply to the demand without changing the total number of pieces in a pre-pack, we will remove one piece of a Flop-Dog-Size from the pre-pack and add one piece of a Top-Dog-Size instead. At the end of the sales period (i.e., at the end of the next delivery period), we can do the TDI-analysis again and adjust accordingly. Given the usual lead times of three months this leads to a heuristics that reacts to changes in the demand for sizes with a time lag of nine month to one year. Not exactly prompt, but we assume the demand for sizes to be more or less constant over longer periods of time. The most interesting question for us was how much, in practice, could be gained by performing only one step of the heuristics explained above. 6. A real-world blind study In this section we describe the set-up and the results of the blind study carried out by our partner. A summary of parameters can be found in Table 4. Test period : April through June 2006 (3 Months) Data collection period: April through September 2006 (6 Months) Branches: 20 branches with unbalanced TDIs randomly classi?ed into 10 test and 10 control branches Pieces of merchandise: approx. 4 000 pieces for all test and control branches Merchandise value: approx. 30 000 e Table 4. Summary of parameters of the blind study

6.1. Selection of branches. A reasonable selection of branches for a test and a control group had to meet essentially three requiremens: ?rst, only those branches should be chosen whose TDI indicated that, in the past, the supply by sizes did not meet the demand for sizes; secondly, no branch should be chosen, where other tests were running during the test period; thirdly, the assignment of branches to test- and control group should be completely random. The reason for the third aspect was that this way all other in?uences on the gross yield than the selection of sizes would appear similarly in both the test group and the control group and, thus, would average out evenly. We suggested a set of 50 branches with interesting Top-Dog-Indices to our partner. Out of these 50 branches, our partner chose 20 branches where a potential re-packing of pre-packs would be possible. This set of 20 branches was ?xed as the set of branches included in our blind study. After that, a random number between 0 and 1 was assigned to each of the 20 branches. The 10 branches with the smallest random numbers were chosen to be the test group, the rest was taken as the control group.

12

¨ ¨ ¨ SASCHA KURZ, JORG RAMBAU, JORG SCHLUCHTERMANN, AND RAINER WOLF

special no test group advertising advertising remove add remove add Branch 1 L XL L XL M XL M XL 2 3 L XL S XL 4 L XL S XL M XL M XL 5 6 M XL M XL M XL M XL 7 8 L XL S XL M XL M XL 9 10 M XL M XL Table 5. How the pre-packs were modi?ed in the test group

6.2. Handling of pre-packs. Next, we had to specify the modi?cations to the size distribution in pre-packs on the basis of the TDI information. It turned out that additional side constraints had to be satis?ed: Whenever a product would appear in an advertisement ?yer, the pre-pack had contained at least one piece in each of the four main sizes S, M, L, and XL. That means, although sometimes the TDI suggested that S was the amplest size, we could not remove the only piece in S from the pre-pack. We removed a piece of the second amplest size (M or L, of which there were two in the unmodi?ed pre-packs) instead. To all branch deliveries, one additional piece of XL was packed, because this was the scarcest size in every branch in the test group. This way, the total number of pieces was unchanged in every pre-pack, as suggested in Section 5. Since all orders had been placed well before the decision to conduct a blind study, our partner re-packed all pre-packs for the test branches according to Table 5. 6.3. Time frame. The test included two relevant time periods: the ?rst period from which the TDI data was extracted and the test period in which the recommendations based on the TDI data were implemented for the test group. The TDI data was drawn from a delivery period of nine month (January through September 2005) and a sales period of twelve month (January through December 2005). The test data was drawn from a delivery period of three months (April through June 2006) and a sales period of six months (April through September 2006). 6.4. Data collection. In order to sort out contaminated data easily, our partner agreed to take stock to check inventory data for correctness every month. To receive a good estimate for the ?nancial bene?t of the supply modi?cation proposed by the procedure described in the previous section we had de?ned some criteria how to detect contaminated data automatically via a computer program. 6.5. Data analysis. The initial situation was given by the data set described in Section 3. In Figure 5 we have depicted the initial Top-Dog-Indices for the test branches and in Figure 6 the initial Top-Dog-Indices for the control branches. Moreover, Figures 7 and 8 ?nally show the situations for the individual branches.

THE TOP-DOG INDEX

13

Top-Dog Index (TDI) for Various Samples (Mean of Test Branches before Field Study)
1,8

1,6

1,4

1,2

1

S M L XL

TDI
0,8

0,6

0,4

0,2

0 Median of all Samples Sample 1 Sample 2 (Complement of 1) Sample 3 Sample 4 (Complement of 3) Sample 5 Sample 6 (Complement of 5) Sample 7 (All Data)

Sample No.

Figure 5. TDI in the test branches from historic sales data The analysis of our ?eld study was intended to answer the following two main questions: are the Top-Dog-Indices better distributed in the test branches than in the control branches, and, if yes, does this have a signi?cant monetary impact? To investigate the latter, we had to analyze some monetary variables. The most important monetary indices for our partner are the gross yield and the last price. The gross yield directly shows how much turnover was lost using a price cutting strategy to sell out all items. The last price tells us how far one was forced to dump items provoked by an inadequate size distribution of the supply. Since the values of di?erent products vary widely we only consider relative values. So, the gross yield is given by the ratio of realized turnover and theoretic revenue without cutting the prices. Since we had to deal with a large amount of lost or inconsistent data, we have applied two ways of evaluating gross yield and last price. Imagine that your data says that you have sold 10 items of a given article in a given branch but that the supply was only 8 items. Or the other way round that the initial supply was 10 items, during the time 8 items were sold, but all items are gone. Both of the described situations occurred signi?cantly often in our data set. Our ?rst strategy to evaluate the given data was to ignore inconsistent data. In the ?rst case, 8 sale transactions are consistent. For the remaining two items the corresponding supply transaction is missing. So, we simply ignore these transactions. In the other case we would ignore the supply of two items. The alternative to ignoring inconsistent data is to estimate or recover it from the rest of the data set. As an example, we would simply assume that there was a supply of 10 items instead of 8 items

14

¨ ¨ ¨ SASCHA KURZ, JORG RAMBAU, JORG SCHLUCHTERMANN, AND RAINER WOLF

Top-Dog Index (TDI) for Various Samples (Mean of Control Branches before Field Study)
1,8

1,6

1,4

1,2

1

S M L XL

TDI
0,8

0,6

0,4

0,2

0 Median of all Samples Sample 1 Sample 2 (Complement of 1) Sample 3 Sample 4 (Complement of 3) Sample 5 Sample 6 (Complement of 5) Sample 7 (All Data)

Sample No.

Figure 6. TDI in the control branches from historic sales data at the same price level in the ?rst case. In the second case we would assume that the remaining 2 items were also sold. Maybe they were shoplifted, some sort of selling for a very cheap prize. So, we need an estimation for the selling prize of the two missing items. Here, we have used the last selling price over all sizes for this product in this branch as an estimate. Neither evaluation method re?ects reality exactly. Our hope was that both estimations encompass the true values. At least our partner accepts both values as a good approximation of reality. The truth may be somewhere in between both values. 6.6. Results. We have depicted the new Top-Dog-Indices after applying our proposed repacking in Figure 9 for the test branches and in Figure 10 for the control branches. Figures 11 and 12 ?nally show the results for the individual branches. We can see in Figure 8 and 12 that the absolute Top-Dog-Indices are not as constant over seasons as we would wish. However, the induced order on sizes seems quite stable accross seasons, and this is all we try to exploit. We can see that it is rather hard to compare the Top-Dog-Indices of the same branches before and after the blind study. The situation on the real market almost never stays constant over time. There are so many in?uences not considered in our study that it would have been a bad idea to measure a possible raise of earnings directly. For this reason, the simultaneous observation of a test group and a control group makes all outer e?ects appear in both. Comparing Figure 9 and Figure 10 based on the same time period, it appears that the Top-DogIndices of the test group have improved more.

THE TOP-DOG INDEX

15

Figure 7. Individual TDIs in test branches from historic sales data More speci?cally, in Figures 11 and 12 we see that in some of the test branches (especially, Test Branches 5 and 6), Size XL is no longer too scarce, while it remains too scarce in other branches. Moreover, on average over all test branches, the Top-Dog-Indices of Sizes S, M, and L are better balanced, while in the control branches the corresponding Top-Dog-Indices di?er. While the former achievement might have been equally possible on the basis of a statistics aggregated over all branches, it seems that the latter result was made possible only by the branch

16

¨ ¨ ¨ SASCHA KURZ, JORG RAMBAU, JORG SCHLUCHTERMANN, AND RAINER WOLF

Figure 8. Individual TDIs in control branches from historic sales data

dependent information from the Top-Dog-Index, since di?erent sizes were removed from the prepacks in favor of XL. Moreover, some of the test branches still need fewer pieces in Size XL, some test branches don’t. That is, in the next optimization step, the branch dependent information becomes vital also for the supply with Size XL.

THE TOP-DOG INDEX

17

Top-Dog Index (TDI) for Various Samples (Mean of Test Branches after Field Study)
1,4

1,2

1

0,8

S M L

TDI
0,6

XL

0,4

0,2

0 Median of all Samples Sample 1 Sample 2 (Complement of 1) Sample 3 Sample 4 (Complement of 3) Sample 5 Sample 6 (Complement of 5) Sample 7 (All Data)

Sample No.

Figure 9. TDI in the test branches in test period But is an improved Top-Dog-Index really an improvement for the business? To answer this question, we have quanti?ed the gross yields and the last prices in the test group and the control group, resp. In Figure 13 we have compared the average values of the gross yield and the last price for the control and the test branches for both evaluation methods. The gross yield of the test branches is 98.0 % using the ignore and 97.2 % using the estimate method. For the control branches we have gross yields of 97.2 % (ignore) and 95.7 % (estimate). This corresponds to improvements of 0.85 and 1.5 percentage points, resp. The improvement for the last price is even larger. The test branches show a last price level of 94.2 % (ignore) and 94.1 %; the control branches exhibit a last price level of 92.5 % (ignore) and 92.7 % (estimate). This corresponds to improvements of 1.7 and 1.4 percentage points, resp. The drastically improved results for the respective “loser branches” (see Figure 14) and the reduced standard deviation (see Figure 15) in the data of the test branches provides evidence for the fact that our procedure was able to reduce the risk of a very low last price or a very low gross yield in an individual branch. This e?ect is desirable beyond the better earnings, as very low last prices undermine the image of the retailer. 6.7. Statistical evidence of the improvement of the gross yield. As in the previous sections we analyze our results concerning the gross yield from the statistical point of view. Faced with the fact of a widely varying gross yield over the branches with no appropriate theoretic sales model, we have to restrict ourselves to distribution-free statistics.

18

¨ ¨ ¨ SASCHA KURZ, JORG RAMBAU, JORG SCHLUCHTERMANN, AND RAINER WOLF

Top-Dog Index (TDI) for Various Samples (Mean of Control Branches after Field Study)
1,4

1,2

1

0,8

S M L

TDI
0,6

XL

0,4

0,2

0 Median of all Samples Sample 1 Sample 2 (Complement of 1) Sample 3 Sample 4 (Complement of 3) Sample 5 Sample 6 (Complement of 5) Sample 7 (All Data)

Sample No.

Figure 10. TDI in the control branches in test period Therefore, we adapt the Wilcoxon rank sum test to our situation. This is a test method to ?nd out whether or not two data sets are drawn from the same distribution. We sort in increasing order the gross yields of the 20 branches participating in our blind study. We associate the largest value with Rank 1, the second largest with Rank 2, and so on. Then we form the rank sums of the test branches and the control branches, resp. The more the rank sums di?er, the less likely is the event that our method did not in?uence the gross yields/last prices at all. For the Wilcoxon rank sum test, it is vital that we have partitioned the 20 branches for the blind study independently at random into test and control branches. It is intuitively clear that a smaller rank sum for the gross yield/last price is more likely if the corresponding expected values are better. A lower rank sum can indeed occur by pure coincidence, but the probability decreases with the rank sum. As an example, the rank sum for the test branches regarding the gross yield measured by the ignore method is 89. If we had not changed anything, the chance to receive a rank sum of 89 or lower would have been 12.4 %. So, we have a certainty of 87.6 % that our proposed re-packing improved the situation. (More formally: the probability that the gross yields of the test branches and the gross yields of the control branches stem from the same distribution, i.e., nothing has changed systematically, is at most 12.4 %.) Now we consider di?erent scenarios. Let yi (b) be the gross yield measured with the method ignore of branch b and ye (b) the gross yield measured with the method estimate. By ic we denote c the scenario where we consider the values yi (b) for control branches and the values yi (b) + 100 for test branches. Similarly, we de?ne the scenarios for ec utilizing ye (b) instead of yi (b). In Table 6 we have given the rank sums and the certainties of some scenarios.

THE TOP-DOG INDEX

19

Figure 11. Individual TDIs in test branches in test period i?0.00 e?0.00 i?0.25 e?0.25 i?0.50 e?0.50 i?0.75 e?0.75 e?1.00 e?1.50 control group 121 131 118 128 105 124 94 120 112 107 test group 89 79 92 82 105 86 116 90 98 103 certainty (%) 87.6 97.4 82.4 95.5 48.5 91.7 19.7 86.0 68.5 54.4 Table 6. Ranks sums and certainties of improvements of the gross yield

20

¨ ¨ ¨ SASCHA KURZ, JORG RAMBAU, JORG SCHLUCHTERMANN, AND RAINER WOLF

Figure 12. Individual TDIs in control branches in test period

How can we interprete these numbers? The ?rst two columns of Table 6 show that with a certainty of 87.6 % (ignore) and 97.4 % (estimate) that our proposed modi?cation increased the expected gross yield. In Scenario i?0.25 we arti?cially decrease the gross yield (ignore) values by 0.25 percentage points. The monetary value associated with this speci?c decrease can be interpreted, e.g., as the implementation and consultancy costs of the modi?cation. So, by a look at Table 6

THE TOP-DOG INDEX

21

100,00%

95,00%

90,00% test branches control branches 85,00%

80,00%

75,00% last price (sorted out) last price (estimated) gross yield (sorted out) gross yield (estimated)

Figure 13. Average last prices/gross yields per merchandise value

100,00%

95,00%

90,00%

test branches control branches 85,00%

80,00%

75,00% last price (sorted out) last price (estimated) gross yield (sorted out) gross yield (estimated)

Figure 14. Minimal last prices/gross yields per merchandise value we can say that with a certainty of 82.4 % our proposed modi?cation yields an improvement of the gross yield (ignore) by at least 0.25 percentage points. 7. Conclusion and outlook The distribution of fashion goods to the branches of a fashion discounter must meet the demand for sizes as accurately as possible. However, in our business case, an estimation of the relative demand for apparel sizes from historic sales data was not possible in a straight-forward way. Our proposal is to use the Top-Dog-Index (TDI), a measure that yields basically ordinal information about what were the scarcest and the amplest sizes in a product group in a historic sales period. This information was utilized to change the size distributions for future deliveries by replacing one piece of the amplest size by a piece of the scarcest size in every pre-pack (this can be seen as a subgradient improvement step in an iterative size distribution heuristics based on the TDI analysis). Empirical evidence from a blind study with twenty branches (ten of them, randomly chosen, were supplied according to TDI-based recommendations; ten of them were supplied as before) showed

22

¨ ¨ ¨ SASCHA KURZ, JORG RAMBAU, JORG SCHLUCHTERMANN, AND RAINER WOLF

10,00% 9,00% 8,00% 7,00% 6,00% 5,00% 4,00% 3,00% 2,00% 1,00% 0,00% last price (sorted out) last price (estimated) gross yield (sorted out) gross yield (estimated)

test branches control branches

Figure 15. Standard deviation of last prices/gross yields per merchandise value a signi?cant increase in gross yield: On average, the increase in the gross yield in our blind study was around one percentage point. The probability that gross yield improvements of at least 0.25 percentage points occurred is at least 87.6 (even 95.5 % if inconsistent data is repaired in a plausible way). And: this was the result of a single iteration of the optimization procedure, which did not result in perfectly balanced Top-Dog-Indices. Given the large economies of scale of a fashion discounter, we consider the TDI a valuable contribution to revenue management tools in this business sector. Moreover, to the best of our knowledge, our blind study is the ?rst published study that evaluates a revenue management method in the apparel retailer industry by comparing simultaneously obtained business results of testbranches (optimized) and control-branches (no action). The draw-back of the TDI is its lack of information about the cardinal expected revenue for a given size distribution of the supply. This is partly due to the fact that the loss of a bad size distribution is closely related to the markdown policy of the discounter. This markdown policy, however, is itself subject of revenue management methods. Therefore, we regard the integration of size and price optimization, as is done in our project BFS-DISPO, as a valuable direction of further research . References
1. Gabriel R. Bitran, Ren? Caldentey, and Mondscheinm Susana V., Coordinating clearance markdown sales of e seasonal products in retails chains, Operations Research 46 (1998), 609–624. 2. L. M. A. Chan, Z. J. Max Shen, David Simchi-Levi, and Julie L. Swann, Coordination of pricing and inventory decisions: A survey and classi?cation, Handbook of Quantitative Supply Chain Analysis: Modeling in the EBusiness Era (David Simchi-Levi, S. D. Wu, and Z. J. Max Shen, eds.), Kluwer Academic Publishers, 2006, pp. 335– 392. 3. Wedad Elmaghraby and Pinar Keskinocak, Dynamic pricing in the presence of inventory considerations: Research overview, current practices, and future directions, Management Science 49 (2003), 1287–1309. 4. A. G¨rhan K¨k and Marshall L. Fisher, Demand estimation and assortment optimization under substitution: u o methodology and application, Preprint FRPS04-135, Duke University, 2006, To appear in Operations Research. 5. Iana Kouris, Dynamic pricing strategies: Markdown procedure for a textile retailer, Diplomarbeit, Universit¨t a Bayreuth, Fakult¨t f¨r Mathematik, Physik und Informatik, 2007. a u 6. Menachem Levanoni, Ying Tat Leung, and Sanjay Elathur Ramaswamy, Method and apparatus suitable for demand forecasting, US Patent 6976001, 2006.

THE TOP-DOG INDEX

23

7. Siddharth Mahajan and Garrett van Ryzin, Stocking retail assortments under dynamic consumer substitution, Operations Research 49 (2001), 334–351. 8. Dawn M. Rose, Steven M. Leven, and Jonathan W. Woo, Assortment decisions, US Patent 7006981, 2006. 9. Christoph Wopperer, Optimierung der Preisreduzierungsstrategie eines Textildiscounters, Diplomarbeit, Universit¨t Bayreuth, Fakult¨t f¨r Mathematik, Physik und Informatik, 2007. a a u

¨ ¨ ¨ Sascha Kurz, Fakultat fur Mathematik, Physik und Informatik, Universitat Bayreuth, Germany E-mail address: sascha.kurz@uni-bayreuth.de ¨ ¨ ¨ ¨ Jorg Rambau, Fakultat fur Mathematik, Physik und Informatik, Universitat Bayreuth, Germany E-mail address: joerg.rambau@uni-bayreuth.de ¨ ¨ ¨ ¨ ¨ Jorg Schluchtermann, Fakultat fur Rechts- und Wirtschaftswissenschaften, Universitat Bayreuth, Germany E-mail address: j.schluechtermann@uni-bayreuth.de ¨ ¨ Rainer Wolf, Betriebswirtschaftliches Forschungszentrum fur Fragen der mittelstandischen Wirtschaft ¨ e. V. an der Universitat Bayreuth, Germany E-mail address: rainer.wolf@uni-bayreuth.de


相关文章:
词汇选择题
agreed ___________ the need for consistency...good-rounded 8. One thing that disturbs me is ...demand C. invoice D. quote 3. late! A. ...
视听阅读book4词汇
agreed ___________ the need for consistency...all-round D. good-rounded 19. One thing that ...demand C. invoice D. quote 44. This was a ...
运营管理习题
it is not necessary to keep consistency between...What is the total cost of the whole year :d ...7. The demand at that location is normally ...
金融英语单选题练习
Consistency D. Going concern E. Money measurement...Demand deposit 4. Risks associated with investing ...The term foreign exchange is best defined by the...
unit3 words
stick 1.Should she feed her child on demand or... are graded for size, shape and consistency . ...of the day to find a puddle of water on top...
第11章 人力资源管理与甄选
High supply/high demand = selection, placement, ...opportunity to get best candidates counters ...consistency of the measurement Commonly Used Tests ...
爱立信无线(江苏赛区)试题和答案
D 20. Which of the following paging strategies ... CLS regards dedicated and on-demand PDCHs as ...1) Adjustment 2) Update 3) Consistency Check 4...
B正式
Consistency D. Going concern E. Money measurement...A. supply of sterling will increase, demand for... New York and Tokyo and the top three ...
人力资源英文版 雷蒙德第七版 (考试重点名词解释)
resides at the top of the organizational chart. ... that accurately predicts future labor demand. ...CHAPTER 5 Reliability: The consistency of a ...
剑桥BEC高级真题第三辑TEST3_图文
assess the likely demand for advanced data ...effectiveness and the consistency of our service. ...20 What would be the best title for the ...
更多相关标签: