Background

Demand planning always has new terms that are necessary to keep up with. Two of the present ones are demand shaping and demand sensing. I decided to include both in an article because for some time there, I would get the two confused for one another, and its important not to do as they are unrelated, except for being two terms which are currently popular.

Demand Shaping

Demand shaping is the process of creating incentives through with customers that smoothes demand, or in eliminating pre-existing incentives such as promotions or end of quarter pushes which distort the demand history making forecasting more difficult to perform.

Demand Sensing

Demand sensing is the use of a procedure to analyze the demand history in order to gain new insight as to how to develop a better forecast, and to make changes in the short term to the forecast. This entry into Wikipedia on the topic is patently ridiculous..

The typical performance of demand sensing systems reduces near-term forecast error by 30% or more compared to traditional time-series forecasting techniques. The jump in forecast accuracy helps companies manage the effects of market volatility and gain the benefits of a demand-driven supply chain, including more efficient operations, increased service levels, and a range of financial benefits including higher revenue, better profit margins, less inventory, better perfect order performance and a shorter cash-to-cash cycle time. Gartner, Inc. insight on demand sensing can be found in its report, “Supply Chain Strategy for Manufacturing Leaders: The Handbook for Becoming Demand Driven.”

This entry was clearly made by a software vendor selling demand sensing software. No forecasting methodology produces a reduction in forecast error by 30%. If the forecast error were 30%, this would mean demand sensing improved the forecast by 30 % x 30 % or 9 percentage points. I would very much like to see the hard data on those numbers. Secondly, what is near term forecasting? Forecasting is produced long term and the sales orders exceed or are short the forecast. I am not sure that “near term” forecasting is a legitimate term. If the lead time is longer than the duration of the short term forecast to sale, how to adjustments to the forecast help?

Searching for Logical Consistency

The more I read on demand sensing, the less it makes any sense. Firstly, many of the things that are often touted as part of demand sensing are should actually be performed by the normal demand planning system. Secondly, making short term changes to the forecast introduces a great deal of noise into the forecast. Forecasts need to be set, and left unchanged. Sales orders may be higher or lower than the forecast, but the forecast is a value which is to bet frozen. The entire purpose of creating a forecast is that there is a lead time, if there were no lead time, there would be no need to forecast. The entire concept of demand sensing is on very weak ground. At this point it seems to be simply a trendy term which is used to sell software rather than any substantial addition to the realm of demand planning. Secondly, in seeing demand sensing put into action at several companies, there tends to be two opinions as to whether the software is adding any value. Those with an incentive to be able to change the forecast love the idea, and those that do not complain of the constant noise demand sensing introduces. One problem I consider is what is the actual forecast that is archived? Is it the original one, the demand sensing adjusted one, one in between? This brings up more questions.

Using Them

Demand shaping is a very valuable function, however extremely few companies actually perform demand shaping. In fact the vast majority of companies perform the opposite or demand distortion. This is because the supply chain department does not control the conditions and terms or pricing that their product is offered to customers under. This is determined by sales or marketing, which generally does not give two hoots about how difficult this makes it for operations to actually fulfill the demand, or what the cost of sales is. Therefore its strange to hear so much about demand shaping, when demand distortion rules the show. So while its an important concept, there is very little chance of this occurring, so it makes not a lot of sense to continue to discuss. The topic should be passed to sales and marketing, who will promptly put the idea in the waste bin.

Conclusion

Neither of these will amount to much in the longer term. One is a great idea, but companies are positively opposed to its implementation, and actively engage in demand distortion, thinking this maximized profits and helps them meet quarterly numbers in a pinch. The other has little foundation in any logic, and seems to distract companies from improving their forecasting capability by adding an post-forecasting processor that jiggers the forecast around. Truly, there are so many opportunities to improve forecasting just in applying truly innovative technologies such as attribute based forecasting,

http://www.scmfocus.com/demandplanning/2010/07/pivot-forecasting/

or discontinuing active forecasting for unforecastable products,

http://www.scmfocus.com/demandplanning/category/pivot-forecasting/

…that its difficult to see how these two concepts are worth the effort to implement.

Background

I was recently reading “The Business Forecasting Deal” by Michael Gilliland. I liked this book for its practical advice on forecasting. I was reviewing the section on middle out forecasting which I have quoted below:

Q: We typically have four to six levels in our hierarchy. What’s the best way to “roll it up”? Top down? Middle out?

A: Usually middle out, from one of the intermediate levels works best. But you will have to determine this from your own data.

Demand at the most granular level of the hierarchy is often difficult to model effectively, because demand is too sparse or erratic. You end up with a bunch of flat line models at the lowest level, representing average demand, failing account for seasonality that truly exists but is undetected amidst all the random noise.

At the highest level of the hierarchy, you miss out on the subtleties of behavior in the lower levels. A retailer might sell swimsuits and snow shovels, both highly seasonal items with very different seasons–and this detail would be lost in a model of overall behavior.

As anyone knows who reads this blog, I believe the future belongs to attribute based forecasting rather than the traditional approach of static hierarchies.

http://www.scmfocus.com/demandplanning/2010/07/pivot-forecasting/

In my upcoming book I explain how in attribute based demand planning systems, the only hierarchy that actually exists is in the user interface, and in the mind of the demand planner. An attribute category, such as color, is a column in a table, and the individual attribute (blue, yellow, orange) is simply selection among options within the attribute category. In most forecast books I have read that deal with this topic, the term top down or middle out is used, because the hierarchies were in fact “hierarchies” in that the levels (what we call attributes) in the hierarchy were not alterable. Region was above Division, Division was above Color and Color was above Size. So a top down forecast would be forecasting at the Region, and a middle out forecast would be forecasting at anything below the top level of the hierarchy but above the actual SKU. The SKU level forecast was then called a bottom up forecast. However with attributes forecasting systems there are only two types of forecast approaches with respect to aggregation. There is a forecast created by an attribute, and there is a forecast which is created without any attribute, or at the SKU level. That is it. In fact, even the term “top down” by an attribute is not actually technically correct, as attribute based forecasting systems have no top. There is simply an association between attribute and a SKU, which is not a hierarchical relationship, but a relational one.

Conclusion

If you use a static hierarchical forecasting system, then top down and middle out forecasting terminology is still valid. However, these systems will become less and less common in the future. Along with it, the terminology which was descriptive of these types of systems is not descriptive of attribute based forecasting systems. This is positive, because attributes are both easier to understand and attribute based forecasting systems are much easier to implement. There is no reason to carry the baggage of a backwards design into to something which is new and requires is own mental models and terminology.

References

“The Business Forecasting Deal: Exposing Myths, Eliminating Bad Practices, Providing Practical Solutions,” Michael Gilliland, (Wiley and SAS Business Series), 2010

The Oracle of Delphi providing advice and forecasts, but without domain expertise.

Background

In this post..

http://www.scmfocus.com/demandplanning/2010/06/forecastable-non-forecastable-formula/

…and in my upcoming book “Statistical and Consensus Demand Planning Software,” I make the point out that a portion of the product database should be identified as unforecastable. In his book, Michael Gilliland makes a similar point when he describes “over-fitting,” which means either developing a custom forecasting model, or using best fit functionality in an application to in a uni-focal manner only attempt to fit the model to the demand history:

But fit to history should not be the sole consideration when choosing your forecasting model. Blindly choosing the best fitting model and assuming it is the most appropriate for forecasting can be a problem in some forecasting packages, or in the misuse of those packages.

Remember again that our objective is to create good forecasts…The key point for selection should not be the fit of the model, but the appropriateness of the model to the nature of behavior you are trying to forecast.

He then uses the example of attempting to predict the results of flipping a coin, which of course cannot actually be predicted.

You could fit a sophisticated model to this pattern. You could even fit the pattern perfectly and project it into the next year. But this is not the right forecast. You would have over fit such a model to the randomness. The proper model in this case is a straight line at 50% heads. Even though its fit to the history is not great and it won’t forecast particularly well, 50% heads is still the most appropriate forecast to use. It will deliver the most accurate and unbiased forecast possible over time.

What I really like about this example is that Michael uses a ridiculous example, the forecasting of the unforecastable, coin flips. He essentially states that a very complex model can be used to perfectly model history of anything, but that modeling history is essentially easy, and the question that is attempting to be answered is the modeling of the future. For difficult to forecast items, the whole point is that a replica of the past will not forecast the future. While Michael’s example is deliberately ridiculous, this is no joke. People attempt to forecast unforecastable things all the time, and to use a lot of math, or smoke and mirrors to cover up the fact that the item of interest is not forecastable. This is one of the reasons why Wall Street employs so many mathematicians. Complicate math is in essence the new mysticism. In order for anything to be considered mystic, is must be incomprehensible but have an aura of legitimacy. This extends to Kings consulting mystics or oracles as to the results of battles, weather, the harvest, etc..

Across many centuries leaders from Athens and Rome would travel to Delphi in modern Greece to consult the Oracles (mystical women) to find out the answers to the future one the most important matters of state. Interestingly we still have a record of some of their predictions or recommendations. Today, going to a Greek mountaintop for projections is considered quaint, and something to be laughed at, and of course we are a scientific society, which is why we go to a magical island for our fake forecasts.

Adjusting for Overfitting

Michael brings up a method that I have used in the past to control for overfitting, which I actually learned from Bill Tonetti at Demand Works. Demand Works Smoothie actually made this very easy to do, because its adjustment to model options is so flexible. This is to perform a forecast for the most recent previous periods, and then compare the fitted model to a naive forecast. I have used this approach several times in testing attributes, to see which would work best to be incorporated in the companies top down forecasting solution. Michael describes this approach below:

A better approach is to withhold some of the most recent historical data, build a model based on the earlier history, and then test the model performance over the holdout sample.

Conclusion

Overfitting is something which is done all the time, and it results in a worse forecast than a moving average or naive forecast. Humans have a natural tendency to think they can control things and predict things that they can’t. One way which this is reinforced is by taking a group of forecasters, say in financial forecasting, and then choosing the top few who beat the market. This happens quite frequently in the financial press, with those that beat the market on any occasion becoming “stars,” or gurus and this is described by William A. Sherdan.

Given that there are thousands of stock market predictors, pure chance guarantees that at least one of them will make what seem to be remarkably accurate calls and attain guru status. Being a market guru, however, is a short-lived honor, because the likelihood of a repeat performance is remote. the odds of making a truly spectacular predictions in any year is one in a thousand, the odds of a repeat performance is one in a million…The eventual fall of a market guru is inevitable.

This means that gurus’ success is due to chance, and this is made apparent by their inability to duplicate their prediction success, however this does not stop the financial press from continuing to create gurus. The financial press simply denies this evaluation, because it simply brings up a new batch of gurus to replace the old.

References

“The Business Forecasting Deal: Exposing Myths, Eliminating Bad Practices, Providing Practical Solutions,” Michael Gilliland, (Wiley and SAS Business Series), 2010

“The Fortune Sellers,” William A. Sheriden, John Wiley & Sons, 1998

http://sacredsites.com/europe/greece/delphi.html

http://en.wikipedia.org/wiki/Delphi

http://en.wikipedia.org/wiki/Famous_Oracular_Statements_from_Delphi

Background

One of the missed opportunities I have discussed in the past is using macro-economic trends in order to adjust forecasts. This type of adjustment would be performed at the top-level of the hierarchy, and in fact could be done for all products, if the product database in question is sensitive to macro-economic factors. Most products are, with their being a wide variation with consumer non-durables being on one end of the spectrum, and housing related industries being on the other.

Where is the Challenge

The challenge is certainly not in the technology side, as long as you have the right application. The best way to apply a top down macro-economic factor is by attribute. For instance, the overall product database can be coded something like the following:

1. Macro-Economic Affected Products
2. Macro-Economic Semi-Affected
3. Macro-Economic Unaffected Products

The macro-economic factor that is used (unemployment, GDP growth, housing starts, etc..) can then be applied to varying degrees in the first two categories, and not at all in the third category. This of course is only one example for demonstration purposes. There could be 10 different categories like this, and multiple macro-economic factors used. For instance if different macro-economic factors were used, the coding could be the following:

1. Unemployment Statistics Applied
2. GDP Growth Applied
3. Housing Starts Applied
4. No Macro-Economic Factor Applied

As soon as this coding is performed, the descriptions added would appear right in the application data view, so that it is easy to apply the increase or decrease though a top down forecast.

The Data

So, the technology exists to easily do this, however, the problem actually lies in getting the right data. This is because the data released by the government has been so adjusted for political purposes that it is no longer a reflection of what is happening in the economy. These are the descriptive statistics that are quoted in news programs that we collectively used to measure the health of the economy and even the fairness of the economy, and they have been helplessly jerry-rigged. A few examples are listed below:

• The government knows that the unemployment rate is a statistic with great implications. Therefore multiple administrations have adjusted the number from its original calculation to make it look as low as possible.
• Inflation controls the cost of living increases that are paid to employees by both the government and private companies. In order to minimize the cost of increases, institutional power (both public and private) has an incentive to use a calculation method that minimizes the inflation rate.

These are just two examples of manipulated statistics, but there are many others. In what appears to be an unending attempt to control public perception, there are few government statistics that have not been “adjusted” over the past five decades. If a forecast adjustment model is based upon false economic statistics, it will not be effective, and will gradually be seen as a non-value add. One wonders how many macro-economic adjustments to forecasts have been abandoned because of this simple fact. Walter J. Williams has the following story which describes this exact thing:

One of my early clients was a large manufacturer of commercial airplanes, who had developed an econometric model for predicting revenue passenger miles. The level of revenue passenger miles was their primary sales forecasting tool, and the model was heavily dependent on the GNP (now GDP) as reported by the Department of Commerce. Suddenly, their model stopped working, and they asked me if I could fix it. I realized the GNP numbers were faulty, corrected them for my client (official reporting was similarly revised a couple of years later) and the model worked again, at least for a while, until GNP methodological changes eventually made the underlying data worthless. Despite minor changes to the system, government reporting has deteriorated sharply in the last decade or so. - Walter J. Williams

This lead Walter to develop ShadowStats.com, which adjusts the government’s statistics to be realistic, rather than to meet political goals. A few examples of the differences between what ShadowStats.com calculates for major government economic statistics and the official government released statistics are listed below:

1. The unemployment rate has been at roughly 9% for some time. However, ShadowStats shows it as roughly 23%. This is due to a number of factors such as workers who leave the workforce because they have been looking for so long they are discouraged. US Government statistics do not count these people. In fact almost all the recent reduction from 9% to 8.5% unemployment was due to people leaving the workforce, yet it was still celebrated in the media.
2. The inflation rate is currently shown as 4% per year, however ShadowStats estimates a full 2.5 percentage points higher, or 6.5%. Furthermore, ShadowStats shows this differentials for years. If a person made $50,000 in 2010, they would need to have made $53,250 in 2011 and $56,711 in 2012 in order for their standard of living to just be maintained. Most companies tend to give out 3% raises, so every year, unless a person is promoted, or moves to a new job, companies pay less for their employees, and a major way they do this is by colluding with the government (though various conservative think tanks such as the Heritage Foundation or the Hoover Institute) in order to keep the real inflation rate hidden.

Conclusion

US Government economic statistics have been so manipulated over time that they are no long a reliable gauge for the state or performance of the economy, and any company which desires to use macro-economic variables will need to pay a source like ShadowStats in order to get numbers that can be used in a causal model to adjust the product forecast.

References

http:www.shadowstats.com

Background

One of the more ridiculous commercials that I can recall is from Firestone, the company which makes tires (the exploding ones that caused the Ford Explorers to roll at high speeds and kill or severely injure 200 of their drivers and passengers from 2000 to 2003). The overall story of this is quite interesting and can be read here.

http://advertising.about.com/od/carrelatednews/a/firestoneadcamp_2.htm

In any case, they recently developed a catch phrase which is “Firestone or nothing at all.” This would seem to imply that given the choice between buying and Firestone tires and having no tires, that the discriminating tire consumer would prefer to not have any tires, which of course would mean the individual could not drive their car. That would be a very discriminating tire consumer as well, although there is no reason for this snobbery, as Firestone does not make highly rated tires.

False Dichotomy

The framework making a decision a foregone conclusion by limiting the options that are provided. This is called “framing” an argument. I came across a textbook case of a false dichotomy in an article about SAP DP. I have listed the quotation below, but kept it anonymous to protect the guilty party.

I have heard feedback from many business users and demand planners that APO DP is complicated. Some think it is too rigid!!  I have heard that it is too enormous and complex.  At least this is not a criticism.  It is ok for something to be enormous and complex, if it solving a complex enterprise problem.  Imagine Exxon Mobil trying to manage the demand for parts through out its complex global supply chain.

What is the alternative?

Ease of entry and view as easy as excel………

I think people are focused more on ease of data management and data entry.  Excel is obvious and you can manipulate everything without moving everywhere. From my working experience, I can say you don’t want to get there – 25 MB spreadsheets and chaotic vlookups.

SAP Makes the Only Demand Planning Application?

It may surprise this writer, but the options available to companies with respect to statistical forecasting do not come down to SAP DP or spreadsheets. Amazing as it may seem, SAP is only one of many demand planning applications to choose from, and as with the Firestone example SAP DP is not even one of the better applications to choose from.

I have been consulting/teaching Demand Planning for the last 15+ years. Definitely APO DP is NOT a complicated tool. The models actually are quite straight forward and based on standard exponential smoothing algorithms. The users should not worry about the backside models but just understand what to use in which business situation. THIS is the gap. We at…..have helped clients narrow the gap and develop standardized training collateral for their working reference.

That is interesting, but I both personally disagree, and receive nothing but negative comments about DP’s complexity from clients that I work with that use DP. Perhaps the fact that the writer has so much experience with the application somewhat reduces the validity of one’s opinion whether an application is complex. The most important audience for whether a tool is complicated is those that have to use it, rather than those that are experts in it.

As for the models being straightforward, they are, however, there is more to a forecasting application than its forecasting models, in fact DP’s forecasting models are generic, in that all vendors have roughly the same models. Except, DP is actually behind other forecasting systems because the forecasting models which use parameters must largely be set by the user, which many forecasting applications have moved beyond at this point.

http://www.scmfocus.com/demandplanning/2011/03/alpha-beta-and-gamma-in-forecasting/

Areas where DP is very uncompetitive include its user interface..

http://www.scmfocus.com/demandplanning/2010/02/why-are-forecasting-interfaces-so-hard-to-design/

..its antiquated associated data technology..

http://www.scmfocus.com/scmbusinessintelligence/2011/05/associative-data-technologies/

..which is one reason it is so poor at forecast aggregation and hierarchies..

http://www.scmfocus.com/demandplanning/2011/03/is-sap-dp-a-good-tool-for-aggregated-forecasting/

..its ability to perform lifecycle planning..

http://www.scmfocus.com/demandplanning/2010/09/the-problem-using-dp-for-lifecycle-planning/

..the fact that major consulting companies recommend it for consensus based forecasting, which it cannot do..

http://www.scmfocus.com/sapplanning/2010/03/27/how-sap-dp-should-not-be-used-for-consensus-based-forecasting/

..the fact that its best fit functionality is so high maintenance that it is typically not used except in the early stages of the project.

http://www.scmfocus.com/demandplanning/2010/07/getting-best-fit-to-work-in-sap/

SAP DP is by far the worst demand planning application I have worked with, and the demand planning applications I have hands-on experience include:

• i2′s (now JDA’s) Demand Planning
• MCA Solutions SPO
• ToolsGroup
• Demand Works Smoothie
• SPSS

SAP DP has a lot of theoretical functionality, however, vanishingly few clients can actually access it, because the maintenance of DP is so high that companies run into trouble just keeping up and running without the system degrading from its basic settings. This is why its total cost of ownership is so high, and so uncompetitive with other offerings.

http://www.scmfocus.com/servicepartsplanning/2011/08/11/what-if-you-paid-nothing-for-sap-software-how-saps-tco-compares-for-service-part-planning/

Finally, and quite importantly, demand planners hate DP’s user interface (one user described to me that the would like to shake up a can of Budweiser and spray it over the screen when the DP interface comes up) , and this is one of the major reasons that every client I have seen that is live with DP, has planners that greatly rely upon spreadsheet. A good player can be judged by how much better they make those around them, and by that measurement, DP is a very bad player, because companies that use DP are not able to progress in their forecasting. The fact that SAP has been able to to continue to sell DP into accounts is a measure of how truly uncompetitive the overall enterprise software market actually is. This is an environment marked by major corrupt consultancies maximizing their consulting revenues by providing false information to clients about software quality and where better products do not have a fair chance to topple poor solutions which have large multi-national brands behind them.

http://www.scmfocus.com/enterprisesoftwarepolicy/2011/11/29/how-efficient-is-the-market-for-enterprise-software/

Conclusion

So firstly, there are many alternatives to DP in the software marketplace, and most can run circles around DP. Certainly, if a person is skilled enough they may be able to get DP to work properly, however, there is not an unlimited amount of time or resources companies are willing to invest in these types of systems. DP has many development flaws that should be addressed.

References

http://en.wikipedia.org/wiki/False_dilemma

Background

Consensus based forecasting are a major focus area that many companies want to implement in the coming years. However, whenever a major initiate is taken, its important to analyze what the history has been with respect to the software type that is to be implemented. While reading “The Fortune Sellers,” a book which takes a very realistic view on forecasting effectiveness in multiple areas such as weather forecasting and financial forecasting. In fact the book does not focus on supply chain planning forecasting, but still has many lessons that can be generalized to the supply chain space. The book has the following to say about research into the improvement that can be expected from using consensus based methods.

Consensus forecasts offer little improvement. Averaging faulty forecasts does not yield a highly accurate prediction. Consensus forecasts are theoretically slightly more accurate than the predictions of individual forecasters by only a few percentage points, due to the average effect that evens out the egregious errors that individual forecasters periodically make. But consensus forecasts are no more likely to predict key turning points in the economy than the individual forecasts on which they are based, and the few extra points of accuracy gained by averaging do not necessarily make them superior to the naive forecast. - William A. Sherden

I will have to do more research on this myself as to why, which will go into a future book on the history of supply chain planning, however, this is consistent with other research into the Delphi method that was designed by the RAND Corporation. On its face, its seems like a strange result, as domain expertise is often distributed across multiple individuals. Also, S&OP, which is a type of consensus based forecasting absolutely has to bring together individuals to make a shared forecast, because no group would allow just operations, or just finance to create the overall forecast. One point of weakness of the research may also be the software that is used. Many companies attempt to perform consensus based forecasting with statistical software, and few focus on reducing bias. Therefore, as implemented, by companies who don’t really focus on a high quality implementation, it is easy to see how consensus based forecasting can show so little improvement.

References

“The Fortune Sellers,” William A. Sheriden, John Wiley & Sons, 1998

Background

One of the lesser discussed topics in demand planning is the use of historical adjustment. In fact if you perform a Google search on the topic, very little comes up. When compared with other forecasting topics, such as error measurement or forecasting methods, it is interesting how lightly the coverage actually is. However, historical adjustments are required by all of the companies that I have worked with in the past, and most companies I have worked with have struggled with performing it in a functional manner. However, as I will describe in this post, much of this is due to selecting solutions which are not designed to perform historical adjustment. It is clear from this analysis that historical adjustment capability should be given much more weight during the software selection process.

Historical adjustment moves demand history to locations where it did not in fact occur.

What is interesting is the many reasons for historical adjustment

1. When products is being transshipped
2. When a product is being moved to a new location (and the demand history must follow along with it)
3. When there is a direct shipment from a plant to a location which is not part of the normal supply network flow (in this case the forecast must be moved from the plant to another location depending upon how the company wants to recognize the sales)

This is just a sampling, there are more reasons that this and they often vary per company.

Star Schemas, MOLAP, Realignment and DP

The vast majority of forecasting systems are based upon either a star schema or are multi dimensional cubes. These data structures have been built up as very fast and therefore desirable, however, they are software approaches to solve a problem regarding latency that is actually obsolete. Secondly, it seems strange how unfamiliar most people are with how poorly these approaches are in historical adjustment. A star schema based system which I use is called SAP DP. DP does not have a good way to display two different “truths” in the system. When something like a product being move to a new location is required, DP must change the hierarchy through something called a realignment. This is where the historical adjustment is then reflected at the new location. However, realignments are very time-consuming and are not done that frequently. This is the problem with star schemas and MOLAP / multi dimensional cube systems, they are extremely inflexible. This is described in this post:

http://www.scmfocus.com/demandplanning/2011/03/forecast-disaggregation-in-smoothie-vs-sap-dp/

Secondly, the requirement companies have is to be able to switching a product between locations flexibly (and to have their demand history follow), and many of the other reasons for moving demand history are similarly transitory and do not work well with systems that require major system activities like realignment every time a historical adjustment is required.

The Flow Through Table

Because of this, most clients I have worked with perform a custom enhancement. Many times this enhancement is some code and a ZTable in SAP. At a few companies I have worked at this is named the ”flow through table”. What this does is moves the forecast to a different location as an intermediate step before the forecast is sent to the supply planning system.

This table above represents a simplified version of this flow through table. However, in reality it is much more complex than this. The single flow through table which I describe is in many cases multiple tables with different conditions.

However, my observation over multiple accounts is that this flow through table ends up being a major maintenance item and most companies end up being not that satisfied with it. This is not to say that for Star Schema and MOLAP systems that a better option has presented itself, but that while necessarily, flow through tables are not particularly satisfying to the companies that employ them.

Unfortunately because the flow through table often is incomplete, the issue with respect to the adjustment can sometimes become supply planning’s issue. This means that time must be spent enabling or disabling the valid location to location combinations in a supply network. In some cases to allow the forecast to flow through between locations, and in other situations to turn off the forecast flow. This is a poor design and extremely undesirable, however, it is what can happen when the issue of historical adjustment is not properly managed demand planning and the negative effects then spill over into supply planning.

Historical Adjustment in Smoothie

Rather than be dependent upon a custom external table or tables, forecasting software should be able to hold all historical adjustments. It should be able to represent in effect multiple realities. That is one “hard” reality which is the actual demand history, and one or more alternative realities, which is the overlay that we want to place on demand history in order to meet future objectives. Thus, the forecasting system should be able to maintain these historical adjustments for the full demand history, and they should be easy to adjust and to move around. Interestingly, a vendor that I have discussed quite a bit on this blog handling historical adjustment quite easily. Demand Works has this to say on the topic of historical adjustment:

Smoothie has the ability to forecast using history that is different that what actually occurred. The Adjusted History measure is where adjustments to history can be made. This is useful if you would like to ignore early demand, or adjust for non-repeatable events.

Background History adjustments are particularly useful for simulating history that may be copied and rescaled from another item, or correcting for unusual events using an approach that will not influence forecast calculations at aggregate levels. - Smoothie Help

Conclusion

It’s interesting that this is a problem which is so pervasive with users of systems like DP, is so easily handled by a forecasting application that so few of my clients use. This inability to adjust demand history efficiently degrades forecast accuracy and is a major headache and time-consuming activity for so many companies, and so few are aware that this effort is unnecessary. It is also amazing how I have been in so many meetings going back to the 1990s where the topic of realignment was discussed. Realignment is completely unnecessary, but consumes a great deal of time and is discussed in supply chain forecasting departments across the country and internationally as well. I wonder how much longer companies will continue to use antiquated approach to manage their forecast history.

References

http://www.scmfocus.com/demandplanning/2011/05/a-better-way-of-importing-data-into-forecasting-and-analytic-systems/

http://www.scmfocus.com/scmbusinessintelligence/2011/05/associative-data-technologies/

Smoothie Help

In this painting, Personification of Astrology from 1650, a woman uses an astrological device to predict the future. Many forecasting techniques through history, including many today are not any more than this. There is an entire industry (actually several) devoted to churning out and charging top dollar for forecasts with no validity.

Background

This blog primarily focuses on supply chain forecasting. However, it is interesting occasionally to observe other areas of forecasting in order to learn from it. Wall Street is an interesting area to observe because the most money in business spent on forecasting is spent in financial forecasting. This industry also routines employs PhD level mathematicians, which is a rarity in industry. They can do this of course because Wall Street pays so well. However, what all of these people and money and systems cannot seem to do is to create forecasts that are better than coin flipping or horoscope reading. The following quotations are interesting passages from a book on forecasting called “The Forecast Sellers.” He beings by describing how anthropologists might look at Wall Street in the future.

To a sociologist from the distant future, the efforts of Wall Street brokers and analysts to predict the stock market at the end of the twentieth century might seem comparable to the behavior of an ancient tribe. Such an observer might describe their behavior as follows: Wall Street was a tribe that inhabited the southern tip of a small island and wielded enormous influence throughout the globe. The tribe worshipped a superior life form called “The Market,” and their lives were completely consumed by contemplation about The Market’s moods, what it thought, and how it reacted to global events. They fretted over whether The Market was depressed, overexcited, acting irrational, or correcting its past mistakes. The tribe derived its power from its professed ability to answer a single question—”What’s The Market going to do?”— posed constantly by people from all over the world, who said the Wall Street tribe billions and billions of dollars to answer this question. One way or another, every member of the tribe was involved n predicting The Market. - William A Sheriden

There are two major forecasting camps on Wall Street. One is technical analysis, which is strangely named as it is primarily chart reading, and the other is fundamental analysis, which is where things like the economic growth rate, and the price earnings ratio of the stock are looked at. It turns out that neither method can beat pure chance, but Sheriden has a particularly amusing take on technical analysis.

Technical analysis is doomed to fail by the statistical fact that stock prices are nearly random; the market’s patterns from the past provide no clue about its future. Not surprisingly, studies conducted by academicians at universities like MIT, Chicago, and Stanford dating as far back as the 1960s have found that the technical theories do not beat the market, especially after deducting transaction fees. It is amazing that technical analysis still exists on Wall Street. One cynical view is that technicians generate higher commissions for brokers because they recommend frequent movement in and out of the market. On this point, Malkiel commented, “The technicians do not help produce yachts for the customers, but they do help generate the trading that provides yachts for the brokers.” - William A Sheriden

In a terms of being able to continually produce forecasts of no value, but continue to receive paying customers Wall Street seem to have no equal. The primary reason for this is that people have a need to believe that there is a way to improve the averages by buying advice. Strangely, the track records of those selling forecasts are often not checked by those buying forecasts. For instance, a forecaster can have a positive return, but the overall market may be returning higher than that. The measurement that is important is if the forecaster can beat the average, and can consistently do this. Some institutions retain their right to charge for fictitious and highly biased forecasts even after they have been shown to have doctored their forecasts to maximize their revenues. A good example of this is Standard & Poor’s which made a big splash in the news by downgrading the US Debt. However, few news outlets explained to their readers and viewers that Standard & Poor’s (along with the other rating agencies Fitch and Moody’s) had been responsible for providing the highest AAA rating to toxic assets that allowed the corrupt carousel of mortgage backed securities to cause the massive financial crisis of 2007/2008. The Wall Street Journal covered the event as if Standard and Poor was some type of authority the credit worthiness of instruments and institutions as can be seen from this quote below:

S&P removed for the first time the triple-A rating the U.S. has held for 70 years, saying the budget deal recently brokered in Washington didn’t do enough to address the gloomy outlook for America’s finances. It downgraded long-term U.S. debt to AA+, a score that ranks below more than a dozen governments’, including Liechtenstein’s, and on par with Belgium’s and New Zealand’s. S&P also put the new grade on “negative outlook,” meaning the U.S. has little chance of regaining the top rating in the near term.Lessons from other countries, such as Canada and Australia, suggest it can take years for a country to win back its AAA rating. At the same time, the economic impact of past downgrades has tended to be larger when multiple firms move to rate a country’s debt as riskier, as opposed to a single firm acting unilaterally.  - Wall Street Journal

Nowhere in the article is it pointed out that Standard & Poor’s had been selling doctored forecasts and ratings to the major investment banks for decades. However a separate quote from the Center For Economic Policy Research (CEPR) did point out what was obvious to people who had followed the rating agencies.

The decision by Standard & Poor’s to downgrade U.S. government debt reflects its own failings as a credit rating agency. It says nothing about the creditworthiness of the U.S. government. Clearly the S&P downgrade was not based on the economics of the country’s debt. S&P has a horrible track record of incompetence in the housing bubble years – they gave Lehman Bros. AAA rating just before its collapse – and the accounting scandals of the stock bubble years. This downgrade should be seen in this light. It is not a serious assessment of the nation’s fiscal condition. – CEPR

However, many more people read the Wall Street Journal article than read the CEPR article. The Wall Street Journal has a vested interest in maintaining the status quo and it’s quite possible that the WSJ receives advertising revenue from the rating agencies. Secondly the WSJ is owned by News Corp (headed by Rupert Murdoch which wanted to push for deficit reduction which meant scaring the public with false concern over the debt ceiling) and this means that they have little interest in pointing out the actual performance of the rating agencies to their readers. Therefore, Standard and Poor’s, Moody’s and Fitch’s continue to sell bad forecasts regardless of their track record. Interestingly, while corrupt news outlets like the WSJ and propose that Standard & Poor’s makes forecasts that are based upon something, its takes a population with zero memory to believe it.

Beating Markets

Much of the advice given out by the enormous number of financial firms selling forecasts is based upon the idea of beating market averages. However, while there is a way to beat markets, the most proven way is to have access to inside information. It is, for obvious reasons, rare for inside information to be sold, as it is more profitable to simply act upon it. In fact there is very good evidence that the overall “averages” of returns of various financial markets is overestimated for the typical investor, because insiders take a large portion of the overall gains. In fact invoking an average is its own form of forecast. Therefore when an average is quoted, it also must be analyzed. Financial advisors essentially use “averages” to mislead common investors as to the returns they can expect to receive.

This is called yield disparity and can be read about here:

http://www.contrarianfocus.com/counterecon/2008/01/11/yield-disparity/

Secondly, what is often happening with respect to finance forecasting is called overfitting by Michael Gilliland:

Anyone can come up with what sounds like a reasonable explanation of past events, but investors don’t need an explanation of why the market did what it did today. Investors need someone to tell them, with a high degree of accuracy, what the market is going to do tomorrow. - Michael Gilliland

Relationship to Supply Chain Forecasting

Wall Street is particularly thick with fake forecasting, but supply chain forecasting also has some dubious methods. There are some methods such as Croston’s which I think demonstrate limited value, but are highly sought after due to the desperation that people have to find an algorithm that can predict lumpy demand. (read more about this at the link below)

http://www.scmfocus.com/demandplanning/2010/07/crostons-vs-smoothie-methods/

The Croston’s method has a lot in common with many highly mathematical models produced by Wall Street. Because highly complex models cannot be understood by the general population, they are assumed to be better than more simple methods. People simply go gaga over the unintelligible. This is one reason the Catholic Church opposed translation of the bible from latin. Unintelligibility is a great desire of those who seek to control others as this following quotation describes quite well.

A century before Galileo, the Englishman William Tyndale was so upset by the foolishness and corruption of the local clergy that he decided to circumvent them and publish a version of the Latin Vulgate Bible in plain English. The Church strongly disapproved, since this would mean that priests would lose their monopoly as interpreters of the word of God. The Archbishop of Canterbury had already declared translation of any part of the bible as heresy punishable by burning.  Tyndale was trapped by an agent and arrested. He was charged as a heretic and burned at the stake. - Dr. David Orrell

Croston’s is one example of unintelligibility, but there are certainly more in the field of supply chain demand planning.

Conclusion

While this blog attempts to describe methods that do work. It is important to occasionally recognize that forecasting in supply chain, as with forecasting in all other fields is also filled with method and approaches that have been historically proven to not be an advantage, but which continue to be used because people have a need to believe that these methods work. The fact that so many forecasting methods through time (astrology, crystal balls, praying to various gods, animal sacrifice), is evidence that there is a deep human need to believe in the power of a method to tell and even in many cases control the future. There is clear subordination to authority here, in that the concept is that other “smart guys,” can predict the future, and you can benefit if you just figure out who they are.

The vast majority of the population up to modern times did not have the opportunity to become exposed to the scientific method (what is a hypothesis, how is evidence gathered, etc..) and that most of the population and most companies still do not practice structured evidence based approaches to determine what is useful to use to forecast and what is fake forecasting. Therefore, forecasts of extremely dubious value continue to be purchased, and most likely always will.

References

“The Future of Everything: The Science of Prediction,” Dr. David Orrell, Basic Books, 2006

“The Fortune Sellers,” William A. Sheriden, John Wiley & Sons, 1998

“The Business Forecasting Deal: Exposing Myths, Eliminating Bad Practices, Providing Practical Solutions,” Michael Gilliland, (Wiley and SAS Business Series), 2010

http://online.wsj.com/article/SB10001424053111903366504576490841235575386.html

http://www.cepr.net/index.php/press-releases/press-releases/statement-on-the-sap-downgrade

Background

In this article, I describe how the standard analysis of the superiority of Japanese manufacturing over the US and other national manufacturers is leaves out many factors which range from Japanese plants being unionized to the emphasis given to finance and marketing versus engineering and manufacturing.

http://www.scmfocus.com/failedsupplychainconcepts/2011/09/why-the-analysis-of-japanese-manufacturing-is-incomplete/

It is interesting that after writing this article, I came across the following quotation in the book “The Fortune Sellers,” which describes how Japanese manufacturers tend to forecast. This book primarily focuses on how forecasts are oversold in many different areas (economics, finance, weather, etc..) versus their actual demonstrated ability to predict the future.

Japanese auto manufacturers’ strategy is to ignore economic forecasts. Stephen Sharf, a columnist specializing in the auto industry, noted that orders to parts suppliers from Japanese auto manufacturers tended to vary by only 2 percent from original forecasts, which he believes has contributed to their success in producing high-quality cars and high profit margins. In contrast, orders from U.S. auto manufacturers are “all over the map,” because they listen to vacillating economic forecasts, and the “result of listening to these seers… is a chaotic condition that continues throughout the year.”

Multi Factors

As I point out in “Why the Analysis of Japanese Manufacturing is Incomplete,” there are many things that Japanese manufacturers do differently, and they cannot be reduced to simply Lean manufacturing techniques. Interestingly, while many consulting companies are offering Japanese manufacturing floor strategy consulting, but upon checking, none offer Japanese forecasting. For some strange reason all of Japanese manufacturing success is attributed to one thing they do differently, and not the many things that they do differently. Imagine for a moment that one took a recipe for very good chocolate chip cookies, but left out most the ingredients, and only focused very intensely in getting the best out of one ingredient, how well would the recipe transfer into a final product? This is what has been done with the analysis of Japanese manufacturing success.

What is Done Differently?

I would like to say that I knew what specifically the Japanese manufacturing companies did differently. (For Lean proponents who are reading this, Japanese manufacturers are not simply waiting for their bin to empty before placing orders, they are ordering based upon a forecast.)  However, the problem is that very little is written on this topic. I did find this quote below from Toyota Supply Chain Management:

For Toyota suppliers, the forecast is fairly consistent from week to week….the strategy at Toyota is to smooth the production schedule.

Smoothing production, or keeping the production stable from week to week is an often described benefit of Japanese manufacturing. Interestingly I also found this quote in the Great Soviet Encyclopedia:

Smooth flow of Socialist production, an important principle for organizing the production process, which presupposes that all the production units of an enterprise or association will systematically fulfill the quotas of the state plan.

However, you can’t make any money consulting in Soviet manufacturing strategies, so of course smoothing the production schedule must be attributed in popular writing to the Japanese. The fact that the Soviets applied smoothing decades before the Japanse is of no consequence to the many authors of Japanese manufacturing books.

Conclusion

It seems that the Japanese “secret” to forecasting is not to allow the finished good forecast to vary the way that American manufacturers tend to. However, it should be understood that this means accepting more inventory (if demand is fluctuating, companies can either vary their production, or vary their inventory). A level or smoothed production schedule means that finished goods inventory is allowed to rise during periods of low demand. This is interesting, because Japanese manufacturers are also known for lower inventories.

References

“The Fortune Sellers,” William A. Sheriden, John Wiley & Sons, 1998

“Toyota Supply Chain Management,” Roy Vasher, McGraw-Hill, 2009

“The Great Soviet Encyclopedia,” Aleksandr Mikhaĭlovich Prokhorov, MacMillan, 1982

http://en.wikipedia.org/wiki/Great_Soviet_Encyclopedia