RBL Volatility - U.S. Air Force

In November 1998, the Air Force Materiel Command Item Management Division asked the Air Force Requirements Team whether or not the Air Force could reduce the number of readiness-based leveling (RBL) changes and frequency of RBL runs. Since RBL's inception, supply personnel perceived RBL levels to be more volatile than the previous leveling technique of the Standard Base Supply System (SBSS) Repair Cycle Demand Level (RCDL). In other words, RBL seems to change more frequently than RCDL. If RBL were unnecessarily volatile, then more assets would be in the redistribution pipeline, which would result in fewer assets on the shelf for the customers and increased transportation cost.

Our first goal in this study was to reduce the number of level changes by eliminating unnecessary, noncost-beneficial level changes, changes where the savings in expected back order (EBO) reduction are not enough to offset the cost of increasing the redistribution pipeline.

Then we sought to reduce the workload associated with RBL runs by reducing their frequency (for example, from quarterly to semiannually) and/or reduce the major command (MAJCOM) and base-level actions to review and load RBL levels.

RBL Push Levels

RBL currently pushes levels at least quarterly. It can and does push levels between quarterly computations for database corrections and high-priority requirements, such as contingency operations. These levels can and do change based on fluctuations in the D200A-computed worldwide recoverable asset requirement, base demand rates, base pipeline time, depot pipelines, and other base factors. It allocates one national stock number (NSN) at a time across all bases, so a change in one base's pipeline data may cause changes in other bases' levels.

RCDL Changes

The RCDL method can change levels as a result of base demand or base pipeline changes only. It is not affected by other bases, the depot pipelines, or the worldwide requirement. RCDL is computed quarterly and is used for recoverable NSNs for which RBL does not allocate levels (primarily two types of RBL-identified problem items), but RCDL levels do not necessarily change from one quarter to the next. In addition, the SBSS has a rule that dampens RCDL level changes. Unless the absolute value of a new RCDL level is greater than the square root of the old RCDL level, the old RCDL remains in effect. For example, an RCDL of two will not change to one or three since that change is not greater than the square root of two.

Analysis

We conducted our analysis by comparing RBL with other alternative leveling policies to identify a means to reduce RBL volatility. The analysis is divided into three parts. Part one describes the methodology and alternatives for reducing the number of RBL level changes; part two documents the results of the analysis; and part three discusses implementation.

Part One

Using four quarters of historical data, we ran RBL in the two quarters in which we do not receive new D200A worldwide requirements data (April and October, hereafter referred to as the off quarters). We compared the changes in levels from the previous quarter's RBL computation to the current RBL set of levels, RCDL, and alternative models. The D200A requirements data are updated semiannually, and RBL uses that updated requirements data in its January and July runs, the months when the September and March D200A cycle results become available. We wanted to reduce volatility in the RBL in the off cycles because these cycles have the least number of RBL input data changes.

We chose to allow RBL to run as it does today in the requirements cycles, January and July, because the latest worldwide requirement becomes available in those 2 months. One of the reasons for RBL to change is a change in the D200A-computed worldwide requirement. In the off cycles, one of the sources of volatility is already reduced since RBL uses the same D200A requirements data from the previous quarter. One of the features of RBL is to ensure the sum of the base levels does not exceed the worldwide requirement. So to ensure worldwide requirements and base-level consistency, we should run RBL for all items in the first and third quarters when we receive new worldwide requirements data. For example, we would not want to allocate only part of the worldwide requirement just to reduce levels volatility.

Causes of RBL Changes. Because RBL changes levels as a result of base pipeline changes and changes in the D200A-computed worldwide requirement, one would expect RBL to change more than RCDL; indeed that is what we found. When comparing the January 2000 RBL Central Leveling Summary (CLS) file to the October 1999 file, RBL is more volatile than RCDL; 9 percent of RBL levels changed compared to 7.4 percent for RCDL. Those statistics include all levels, such as zero levels. The percentages are higher when only positive levels are measured.

Table 1 shows the causes of RBL changes by case (each case being a stock record account number (SRAN)-NSN combination as displayed in the CLS file). The table summarizes all level changes that occurred in the period from April 1998 to January 1999. Many of the changes resulted from worldwide requirements changes (4.5 percent exclusively, plus some portion of 62 percent). To ensure consistency and accurately allocate the entire worldwide requirement, we applied level reduction alternative changes only in the off-cycle quarters.

Volatility Reduction Techniques. We looked at two different approaches to reducing volatility. One is to run RBL only semiannually, in January and July; in other words, reduce the frequency of runs from quarterly to semiannually. The other approach is to run RBL only for certain items in the off cycles, those for which there is a benefit to changing the levels. For example, for level changes between April 1998 and January 1999, we looked at the cases that had level changes and Doted the EBO reduction.

Table 2 shows that 38.6 percent (21.8 + 16.8) of the RBL levels that changed from April 1998 to January 1999 resulted in less than a .1 EBO reduction. Almost 50 percent had no reduction, or less than .2, and 57 percent had no reduction, or less than .3. So there is little benefit in terms of EBO reduction to changing levels for many of the items that had levels changed in that period. However, some items had level changes that resulted in significant EBO reductions. More than 20 percent had EBO reductions greater than one. New RBL levels should be computed for those items with significant changes.

How do we decide which NSNs RBL should relevel and which changes are significant? How do we decide what amount of EBO reduction is trivial?

NSN Releveling Techniques. We tried two methods to decide which NSNs RBL should relevel because the changes in pipeline data were significant.

Method 1. The first method, RCDL change, would let RCDL determine which items to relevel. In method one, RBL recomputed and, if appropriate, changed levels for those items that the RCDL technique would change in the off-cycle quarter. (As previously noted, we recommend RBL be recomputed for all NSNs in the requirements quarters.) We tested two RCDL techniques, unmodified RCDL and the square root RCDL (the damping rule described earlier that is currently in use) to identify which items RBL should relevel. The advantages of using RCDL are that it changes levels using a technique designed to optimize base levels as a result of changes in base pipeline data, the only data that changes in the off cycles. Also, it did not require major format or program alterations to the RBL data input programs because the RBL model does not require any additional data.

Method 2. The second method, EBO change, required a change to RBL input data; RBL would need to know the previous quarter's level. For this method, RBL relevels items with sufficient EBO reduction. As stated, method two would require changes to the RBL input file and requires RBL to compare last quarter's levels allocation and the resulting EBOs to the current quarter's levels and EBOs to decide if there is sufficient EBO reduction to warrant a level change.

Whether method one or two is applied, RBL would run every quarter; however, in the off-cycles, RBL would push fewer levels, thus reducing variability.

Part Two

Comparison of Off-Cycle Leveling Techniques. Table 3 provides the results of the various methods using October 1998 data.

Table 3 compares the total EBO and number of levels selected for releveling for five alternatives. The number of levels selected does not mean the levels all change; it shows the number of NSN-SRAN combinations identified for RBL to compute. The first alternative, no RBL, uses the previous quarter's levels in the new quarter. (That is, RBL would not run at all in the off quarter.) There is a 53 percent ([10208-6735]/6735) increase in the number of EBOs compared to running RBL for all items, full RBL. Clearly, not running RBL at all is a poor option.

Next, we looked at running RBL on a relatively smaller group of items to achieve most of the EBO reduction without generating new levels for all NSN-SRAN combinations. For other methods, the resultant EBO is close to a full RBL run, and there were fewer cases selected for releveling. Table 4 compares alternative techniques for running RBL in more detail.

RCDL Technique. Table 4 shows that running RBL only fox NSNs that had an RCDL change would not generate any changes for 67,477 NSNs. Full RBL would have releveled 2,844 of those NSNs and achieved an additional 101 reduction in total system-wide EBOs (the sum of all back orders generated by all NSNs leveled by RBL). These 2,844 NSNs show a potential error in using the RCDL method; these additional NSNs should be releveled because they reduce EBOs but were not selected by the RCDL method. The RCDL method identified 35,113 NSNs for a possible level change. RBL would not have changed levels for 20,988 of those NSNs identified for change because basically no reduction in EBOs occurred by altering levels.

Square Root RCDL Technique. The square root RCDL method had similar results. Modified RBL did not relevel 70,411 NSNs because they did not meet the square root RCDL criteria. Full RBL would have releveled 3,992 of those 70,411 NSNs and reduced total system-wide EBOs by 280. Just as in the ordinary RCDL technique, these items again reflect potential error from this method. The square root RCDL method identified 32,179 NSNs for RBL to relevel, of which full RBL would not relevel 19,212.

EBO Technique. The final method identified in Table 4 would relevel items only if there was a change in the EBO of at least 0.1. The EBO (0.1) method identified 71,826 NSNs (all within plus or minus 0.1 EBO) that would not relevel with a total system-wide increase of 27.3 EBOs. Therefore, the EBO (0.1) method identified captured all NSNs for releveling which would generate a reduction in total system EBOs. The EBO (0.1) method identified 30,764 NSNs that would change, although not all SRAN-NSN level combinations would change. The small Increase in system-wide EBOs and prevention of trivial changes roves that the EBO (0.1) method is in practice superior to the RCDL method.

Preliminary Findings. Using the EBO method is also theoretically superior to the RCDL method. With the EBO method, RBL would actually measure the EBO impact of not changing levels as a criterion to select levels for change. With he RCDL methods, modified RBL would decide which NSNs to relevel without taking into account the impact on the requirements system as a whole (increase or decrease in total system-wide EBOs). Therefore, the RCDL methods could still make trivial changes, releveling NSNs while achieving little reduction in system-wide EBOs. Using the RCDL method, modified RBL would not make changes for NSNs that probably should be changed. The number of these errors would be small with the RCDL method, but they would occur. So we selected EBO as the means to identify NSNs for releveling.

EBO Threshold. The EBO method is theoretically and actually superior to the RCDL method; about the same number of level changes occurred with fewer EBOs. What EBO threshold should the Air Force use to select NSNs for releveling? In other words, which changes should be considered as trivial changes?

To answer that question, we ran RBL to see the EBO impact if we added a 10-day RDO pipeline to any base that had a level change. If levels change, theoretically, that means an asset must be redistributed from the base with the decreased RBL to another base with an increased level. We measured the EBO increase caused by the added 10-day RDO pipeline, 10 days being the average time it takes for completion of the RDO process. That EBO increase can be considered the cost of a change in levels. The EBO change threshold, the measure of the benefit achieved by changing a level, should be greater than the EBO cost.

Table 5 shows the frequency of the EBO increase as a result of adding the 10-day RDO pipeline. Excluding outliers, changing levels and adding a 10-day RDO pipeline incurs an average 0.08 increase in EBOs. Using 0.08 as the threshold criteria would exclude 80.96 percent of the level changes. We proposed using 0.08 as the threshold criteria. Basically, if the EBO reduction for the level change is not sufficient to offset the 0.08 EBO increase caused by the RDO pipeline, do not change the level.

Using the EBO Technique. With a method to select NSNs for releveling in the off cycles, EBO, and a threshold criterion (0.08), there are still two issues to consider. Should RBL relevel for new users in the off cycles? What about levels to support PDM account needs?

New Users. In RBL, a change in one level usually is offset by a change in another level, since RBL has a fixed amount of worldwide requirements to allocate. Also, RBL currently determines when a base will receive a positive level. (RBL can and does allocate positive levels to bases with only one demand, and it can and does allocate a zero level to bases with two or more demands.) Should RBL change levels for new users in the off cycles? For bases with a new adjusted stock level (ASL) or with sufficient demand history to receive a positive level, should RBL allocate to that new user regardless of the EBO impact?

Theoretically, RBL should honor a new ASL once it is approved. It is possible a new ASL will cause an increase in the worldwide requirement and, therefore, not affect any base levels except the base with the new ASL. However, it is not clear that RBL should provide a positive level to a new demand user unless the EBO criterion (0.08) is met. Theoretically, pushing a positive level to a new user will incur an added redistribution pipeline. We tested three methods of leveling to try to answer these questions. Table 6 displays the results of using these three methods: the EBO-only rule which ignores new ASLs or new users; releveling using EBO and new ASLs; and finally, releveling using EBO plus allocating levels for new users and ASLs.

Running RBL for all items generated 81,563 level changes. Using the EBO (0.08) method only generated 9,341 changes, 11.45 percent of the full RBL changes.

Using the EBO (0.08) method and honoring all new ASLs generated 16,935 changes. (The Air Force Communications Agency implemented its new stockage policy at this time, which explains the relatively high number of ASL changes.)

If new users were exempted from the EBO (0.08) rule, there would be 69,632 level changes representing 85 percent of the changes that would occur from running full RBL. This, in essence, would negate the benefit of the EBO (0.08) rule and create almost as many level changes as are generated today. We proposed using the EBO (0.08) method and honoring new ASLs, while only releveling for new demand users if there is an EBO reduction greater than 0.08.

PDM. Should the Air Force apply the EBO (0.08) rule to the D035K depot account (PDM) levels? The PDM worldwide requirement is not based solely on the D035K reported daily demand rate. For example, the depot may repair an end item or exchange for use on another item in next quarter's repair cycle. The other item may not have been repaired last quarter or even last year, which means there may not be any DDR for it in the D035K database. So, RBL's measure of EBOs (based on the reported D035K DDR) may not accurately forecast the change in EBOs in this example. Many PDM items are exclusively used at the depot; therefore, there is no redistribution pipeline. For these reasons, we proposed running RBL without constraints for all NSNs in use at the PDM/DO35K accounts in the off cycles.

Conclusions

We proposed running full RBL twice annually to coincide with the D200A requirements cycle. The January RBL run uses September D200A cycle requirements data, and the July RBL run uses March D200A requirements data. Off-quarter (April and October) RBL computations would relevel significant EBO changes as well as ASL and D035K account changes. This results in the relatively lowest total system-wide EBO increase with the fewest number of level changes. Table 7 shows the expected results of the proposed policy.

As indicated in Table 7, RBL currently pushes to the average base, excluding depot retail accounts, more than 1,400 XCA data images (levels). In the off quarters, RBL will only push XCAs for levels that change based on our proposed criteria. For the April and October RBL pushes, bases should average only 100 to 150 XCAs. Our proposal reduced the number of levels pushed in those quarters by 90 percent (1250/1400) and eliminated trivial level changes, those with an EBO reduction of 0.08 or less.

Part Three

Implementation Plan. We briefed our proposal to the Air Force Supply Executive Board, and they approved the changes in February 2000. The Requirements Team, together with the Air Force Materiel Command, fully implemented the changes in October 2000. The RBL model was reprogrammed to push XCAs for EBO changes greater than 0.08 plus ASLs and D035K account changes for the off-cycle RBL runs.

The proposal requires some changes to programs other than the RBL model. AFMC changed the RBL input file to provide the previous quarter's levels. RBL will push only XCAs on levels that change. However, the CLS file (RBL output file) remains a complete file. The CLS should reflect all levels, even those not changed from the previous quarter. This will facilitate accurate item management at the air logistics centers and MAJCOMs.

The Standard Systems Group (SSG) made a programming change to the SBSS. Currently, the SBSS follows up if it does not receive an XCA every 120 days. SSG changed the followup time period to 210 days.

The proposal did not affect out-of-cycle RBL runs and changes. Out-of-cycle, Air Mobility Command forward-supply location level changes will still be honored (XE4 data image with an I procedure) without any leveling change constraints. If the item manager identifies an NSN for RBL to rerun (for example, correcting RBL-identified problem items), RBL will run the item without the leveling change constraints. These out-of-cycle changes must also be posted to the CLS. AFMC has plans to post level changes to the RBL web site and the CLS.

Captain Spencer is a logistics career-broadening officer at the Oklahoma Air Logistics Center, Tinker AFB, Oklahoma. At the time of the writing of this article, he was Chief Requirements Policy Branch, Air Force Logistics Management Agency.

Table 1.
Causes of RBL Changes
(Cases With Positive RBL Only)
Causes of RBL Changes
(Totals from April 1998 to
January 1999)                      Cases Affected
Change in Daily Demand Rate (DDR)           0.46%      (502)
Change in Requirement (Req)                 4.49%    (4,829)
Change in Pipeline                         11.03%   (11,840)
Combination of DDR and Pipeline            20.03%   (21,502)
Combination of DDR, Req, and
 Pipeline                                  62.23%   (66,815)
Adjusted Stock Levels                       1.05%    (1,123)
Other
Total                                     100.00%  (107,368)
Table 2.
Frequency Chart of Level Changes
for April 1998 to January1999
Changes in EBO (Level Changes Only)
   EBO Changes     Number of Cases  Percent of Cases
       0.0                  23,405              21.8
       0.1                  18,053              16.8
       0.2                  11,862              11.0
       0.3                   8,171               7.6
       0.4                   6,024               5.6
       0.5                   4,685               4.4
       0.6                   3,683               3.4
       0.7                   3,060               2.9
       0.8                   2,507               2.3
       0.9                   2,115               2.0
       1.0                   1,673               1.6
[greater than]1.0           22,091              20.6
Table 3. RBL Off-Cycle Run Options
RBL Off-Cycle Run Options
(October 1998)
                       Levels Selected for
Option      Total EBO      Releveling
No RBL       10,208           None
Full RBL      6,375          535,800
RCDL          6,796           62,775
RCDL Sq Rt    6,869           62,775
EBO (0.1)     6,810           64,833
Table 4. Comparison of Alternative Techniques (October 1998 Data)
Comparison of Alternative Techniques (October 1998 Data)
           Unchanged  Additional Changed NSNs Using
Technique    NSNs        Full RBL/EBO Reduction      Changes NSNs
RCDL        67,477             2,844/101.2              35,113
SQRT RCDL   70,411             3,992/280.0              32,179
EBO (0.1)   71,826              4,504/27.3              30,764
           NSNs Full RBL Would
Technique   Not Have Changed
RCDL             20,998
SQRT RCDL        19,212
EBO (0.1)           -
Table 5.
Frequency Chart of EBO Change
EBO Change for 10-Day RDO Pipeline
EBO                  Number   Percent of     Cumulative
Change              of Cases    Cases     Percent of Cases
0.00                 3,002      33.29          33.29
0.01                 1,806      20.03          53.32
0.02                   745       8.26          61.58
0.03                   547       6.07          67.64
0.04                   366       4.06          71.70
0.05                   313       3.47          75.17
0.06                   223       2.47          77.64
0.07                   167       1.85          79.50
0.08 [*]               132       1.46          80.96
0.09                   131       1.45          82.41
0.10                   419       4.65          87.06
0.20                   463       5.13          92.19
0.40                   223       2.47          94.67
0.60                   116       1.29          95.95
0.80                    64       0.70          96.66
1.00                    30       0.33          96.99
[greater than]1.00     271       3.01         100.00
(*)Average increase in EBOS = 0.08
Table 6
EBO Change Analysis
Comparison of Leveling Techniques for New Users
Technique                                   EBO     EBO      Level
                                                  Increase  Changed
Pushing Last Quarter Requirement           6,702      N/A   81,563
Pushing Last Quarter Levels                9,068   26.10%        0
EBO [greater than] 0.08 only               8,162   21.78%    9,341
EBO [greater than] 0.08 w/ASL rule         8,083   20.61%   16,935
EBO [greater than] 0.08 w/ASL & user rule  6,739    0.55%   69,632
Technique                                  Percent Changed (Levels
                                               Changed/8l,563)
Pushing Last Quarter Requirement                      N/A
Pushing Last Quarter Levels                             0
EBO [greater than] 0.08 only                        11.45
EBO [greater than] 0.08 w/ASL rule                  20.76
EBO [greater than] 0.08 w/ASL & user rule           85.37
Table 7
Average Levels Pushed Per Base
Comparison of Levels Pushed (XCAs Released
                        Average Number  Average number
                          XCAs Using      XCAs Using
 Technique               Oct 98 Data     Apr 99 Data
Present Rules
(Full RBL)                  1,471           1,403
EBO [greater than] .08
w/ASL Rule                    153              95

Function of RBL

RBL allocates the worldwide recoverable item requirement to bases and depot accounts (program depot maintenance [PDM]) so as to minimize EBOs for base-level customers and, therefore, the Air Force as a whole. It is an optimization model that uses marginal analysis to allocate the next level to a base or depot that will result in the greatest EBO decrease. Even a small decrease in EBOs resulted in a level change, due to coded algorithms in the RBL model. RBL did not look at the current base level, the previous quarter's RBL allocation, in determining the newest allocation. As a result, RBL changed a level to achieve a reduction of even 0.0001 of an EBO. It also did not consider the cost of the increased redistribution order (RDO) pipeline resulting from these changes and, as a result, potentially placed assets in the RDO pipeline, thus making them unavailable for use. Therefore, potential back orders could have resulted from redistribution of assets among bases to meet the new, changed RBL allocation.

COPYRIGHT 2001 U.S. Air Force, Logistics Management Agency
COPYRIGHT 2004 Gale Group