Forecasting fashion apparel (Part 3)

Mike Gilliland
Mike Gilliland|February 7, 2012

Some ideas sound great (combining chocolate with peanut butter) and turn out great (Reese's Peanut Butter Cup).  Some ideas sound great (getting a face lift) but turn out bad (Kenny Rogers, Greta Van Susteren). Some ideas sound bad (a Run-DMC / Aerosmith duet) but turn out great ("Walk This Way").   Some ideas sound bad (letting your daughter go camping with Levi Johnston) and turn out bad (do I really need to explain this one?). And then there are those ideas that sound terrible to begin with, and turn out even worse.  Let me describe such an idea below.

Forecasting Fashion Colors - The Wavelength Time-Series Approach

When it comes to forecasting, I'm a big fan of creativity and new ideas -- the crazier the better. After failing to crack the fashion color forecasting problem while working in the apparel industry, I got to think more about the issue with the clearer head of an outsider. Is there a way to predict, a year or more in advance, what will be the "fashion" colors of future seasons? And if so, couldn't I corner the market on the dyes and fabrics, and own the fashion apparel market?

My grand idea was this:

I shall convert each fashion color from past seasons into a number -- its wavelength. Then I'll apply the usual time-series forecasting methods to this time-series of wavelengths, to determine the wavelengths of fashion colors for future seasons. 

An idea like this might sound good after a few mushrooms, but I can't use that excuse since I haven't done drugs since 1970.  The more I thought about it, the worse the idea sounded, for a number of reasons:

  • Insider information, of what colors the barons of the fashion industry are planning to push on us, is probably the best way to get the color forecast for next year. Yet I lacked a ticket to Paris or Milan, or access to their private discussions.
  • There is not necessarily a single fashion color in a season, but there may be many. (For example, the Fall 2010 colors were Camel, Rich Purple, Modern Metallic, Mixed Olives, Pale Blue, and Bright Red.)
  • How do I convert the name of the fashion color to a wavelength? For the regular scientific color names, I know Violet has a wavelenth of 400 nm, that Blue is 475 nm, Green is 510 nm, etc. But the fashionistas come up with all kinds of crazy names for their colors, often conjuring up images with inanimate (or animate) objects.  Just in the reds we have Bright Red, Soft Red, and Medium Red, but also Fire Engine, Delicious Apple, Burning Ember, and Rusty Trombone.

The Results

Difficulties aside, I took my best shot at converting the past 25 years of fashion reds into their wavelengths, ran them through SAS Forecast Server, and came up with my forecast for Fall 2012 as 737 nm.  I then went to the Pantone color wheel and realized that 737 nm is off the charts -- it is not a visible color.  So either my method is wrong, or we are going to see a lot of people walking around this fall in the emperor's new clothes.

-------------------------

Wavelength image from Universe by Freedman and Kaufmann.

Post a Comment

Forecasting fashion apparel (Part 2)

Mike Gilliland
Mike Gilliland|February 3, 2012

Have you noticed the annoying stock art they put on The BFD blog header? All I can think of is "If those idiots only used SAS Forecast Server, they wouldn't have to draw graphs all over their window panes just to do forecasting." It must really p.o. the housekeeping staff at that company.

Forecasting Fashion Colors

Last week's The BFD ended with the cliffhanging question of whether we can accurately forecast fashion colors. One way to interpret this question is whether we can, in general, forecast new products.  (While the underlying item itself (like a blouse, or jacket, or refrigerator) may not be new, it is coming out in a new "fashion" color that may only be available for a limited time.)

New product forecasting methods are generally fair at best, and often quite poor. For something in a fashion color, we at least have historical information on the "basic" color versions of the product, and previously released fashion color versions.

Using a structured analogy approach (as described in this whitepaper "New Product Forecasting Using Structured Analogies"), you can visualize the range of demand for analogous new product introductions (see below), giving yourself a ballpark idea of what forecast may be "reasonable."  The structured analogy approach is probably most valuable, however, as a way to detect the unreasonableness in new product forecasts generated by other means.

(Image shows the first 16 weeks of sales for DVDs with attributes: Rating=R, Genre=Horror)

Rather than bet the company on a new product forecast that is probably going to be wrong (and wrong in a big way), we are better off focusing on flexibility and responsiveness of the supply chain. Or at least, in our financial analysis / justification for the new product, accounting for the fact that we really don't know how well it is going to sell. I've given up hope for a miracle new product forecasting algorithm.

In the next installment we'll examine a new theory: The Wavelength Time-Series Approach to forecasting fashion colors.

Post a Comment

Forecasting fashion apparel

Mike Gilliland
Mike Gilliland|January 27, 2012

Ten years ago I spent some time in women's undergarments*, as Director of Forecasting at Sara Lee Intimate Apparel (now Hanesbrands).  Sure, it sounds glamorous -- product posters on our office walls, quarterly runway shows of new products, and partying with the full-figured Playtex models (some of whom were fuller than I figured).  All my former colleagues were jealous I got the job. But believe this, it was difficult and stressful for me and my department.

Whenever things got too intense, my boss Cecil Moore used to cheer us up with this reminder:

Our products don't kill anybody. Sure, they may cause some discomfort and chaffing, or even minor disfigurement when an underwire pokes through. But hey -- they don't kill anybody.

Fine encouragement like that gave us the will to persevere.

Forecasting Fashion Colors

In the apparel business, products can be crudely classified as either "basic" or "fashion." Basic items, like black dress socks or white briefs for men, tend to have reasonably well-behaved and repetitive sales patterns over their lifecycle -- which is expected to be years.  There is little if any technological change in the products, they look and function the same year after year, and may be promoted at the same times every year reinforcing the repetitive pattern.  We can often forecast these items reasonably well, and manage their supply chains without a lot of grief.

Fashion products, on the other hand, are designed to reflect (or define) current tastes. There is no expectation of an ongoing product lifecycle -- they are meant to be built and sold within a single season.  If the item proves popular and sells out, there may be no ability to replenish the supply.  And if too many of the items are made, they get marked-down at season end for closeout sales, moved to alternative sales channels, or simply donated or destroyed.

Often a fashion item is just a basic item in a different color.  Instead of only white, black, and beige (the standard basics in women's intimates), we might offer ten or more additional "fashion" colors throughout the year, and therein lies the problem.

You can imagine the forecasting challenge -- trying to figure out the popularity of each color months in advance.  And you can imagine the financial risk in each offering -- make too few and we lose sales/annoy customers, make too many and we are stuck with excess (and perhaps unsellable) inventory.

A sure way to significantly improve profitability would be to have highly accurate forecasts of fashion colors, but can it be done?  We'll explore this a little deeper next time...

-----------------------

*Probably more time than Rudy Giuliani, but less time than J. Edgar Hoover.

Post a Comment

For the love of forecasting

Mike Gilliland
Mike Gilliland|January 25, 2012

Love can make a person do bad, dangerous, stupid, and irresponsible things.  Love of country can make a politician stray from his wife. Love of Pepsi can make a pop musician lose his hair in a pyrotechnics-gone-bad commercial. Love of acting can make academy award winners accept starring roles in Ishtar. And for me last Wednesday morning, my love of forecasting was so overwhelming and distracting, I forgot to put on shoes and arrived at work in my bedroom slippers.

Beware the "Inside View"

The November 2011 McKinsey Quarterly contains an excerpt from Daniel Kahneman's new book Thinking, Fast and Slow.  Kahneman tells the story of textbook writing project, in which his survey of the co-authors estimated two years (+/- six months) for completion.

Kahneman then asked one of the co-authors if he could think of similar projects, and how long they took to completion.  Here the answer was quite different -- 40% of them failed to finish, and the successful ones had taken seven to 10 years to complete.

At this realization, the more rational course of action would have been to quit.  "None of us was willing to invest six more years in a project with a 40 percent chance of failure." Yet the project continued and the book was completed in eight years -- but was never used.  By then the enthusiasm for the idea had waned.

Kahneman uses this story to illustrate the perils of the "inside view" -- simply extrapolating from our specific circumstances and the information in front of us. This narrow focus on what we know fails to allow for the "unknown unknowns" -- all the crises, unanticipated, and random events that get in the way of our progress. "There are many ways for any plan to fail, and although most of them are too improbable to be anticipated, the likelihood that something will go wrong in a big project is high."

The "outside view" is based on the category or reference class relevant to our prediction.  Here, it was the class of similar book projects.  Kahneman argues that this outside view provides "...a reasonable basis for a baseline prediction: the prediction you make about a case if you know nothing except the category to which it belongs.  This should be the anchor for further adjustments." This estimate can then be adjusted given any case-specific information.

New Product Forecasting by Analogy

Knowingly or not, we employ the outside view when we forecast a new product based on sales of similar products. Such "forecasting by analogy" is perhaps the second most common new product forecasting method (second only to "management forecasting whatever they darn well please").  As Kahneman asserts, "...if the reference class is properly chosen, the outside view will give an indication of where the ballpark is. It may suggest, as it did in our [the book project] case, that the inside-view forecasts are not even close."

Selecting an appropriate reference class is the key step in forecasting by analogy, and is often poorly done. Through mischief or incompetence, we are wont to cherry pick the analogous products (e.g., only the most successful ones) in order to justify the new product forecast we desire.

Find more thorough discussion of this topic in the whitepaper "New Product Forecasting Using Structured Analogies" (available for free download).

Post a Comment

Guest Blogger: Len Tashman previews Winter 2012 issue of Foresight

Mike Gilliland
Mike Gilliland|January 18, 2012

The Winter 2012 issue of Foresight is now available. Here is Editor Len Tashman's preview:

---------------------------

Our last two issues featured Steve Morlidge’s Guiding Principles for managing an organization’s forecasting process. You can see the summary table of these principles on page 31. With this issue, we continue their development by presenting commentaries from researchers, consultants, and practitioners who have devoted their professional careers to the forecasting function. Alec Finney sums it up perfectly:

Steve Morlidge has produced a practical handbook to Scott Armstrong’s encyclopedia of forecasting principles. Steve’s five key themes are just that – the areas that every forecaster and planner needs to get right to deliver actionable information
for decision makers.

The advances in technology through broadband, computational capabilities, and the Internet open possibilities for virtually immediate analysis and interpretation of streamed data. In his article Stream Analytics for Forecasting, Patrick McSharry shares with us his perspectives on the forecasting challenges and opportunities from data streaming.

Our section on Forecasting Methods presents Scott Parrott and colleagues’ article on Forecasting Rounds of Golf. The golf industry has been under financial stress due to local, national, and global recessions and the consequent decline in tourism at vacation areas. As a result, it is paying more attention to improved forecasting methodology, including statistical modeling and judgmental input. Using data from a Myrtle Beach golf course, the authors  combine a statistical model of the trend and seasonal patterns in golf demand with the application of the analytic hierarchy process to extract the judgmental forecast adjustments of golf-course experts. It’s an interesting case study.

The U.S. Presidential Election

Election forecasting will be alive and well in 2012. Many websites now provide up-to-date campaign, polling, and bellwether information and show how the data extrapolate to the November elections. One of the first of these was the Pollyvote. Now the originators of this website share with us The PollyVote’s Year-Ahead Forecast of the 2012 U.S. Presidential Election.

Does the Presidential Candidate’s Campaign Affect the Election Outcome? Successful election forecasters and scholars (such as historian Allan Lichtman) have dismissed the importance of campaigning in the outcomes of U.S. presidential elections, believing instead that it is basically the incumbent’s record of performance that determines his or her party’s chances of retaining the White House. Richard Nadeau and Michael Lewis-Beck report that their reading of the forecasting models indicates otherwise: namely that, in a close election, the quality of the campaign can make a telling difference. And the November election is shaping up to be quite close indeed.

The winter 2012 issue concludes with a glimpse at Daniel Altman's new book, Outrageous Fortunes: The Twelve Surprising Trends That Will Reshape the Global Economy.

---------------------------

Join the International Institute of Forecasters to receive a free subscription to Foresight, as well as to the International Journal of Forecasting and the Oracle of the IIF newsletter.

Post a Comment

Myers-Briggs Type Indicator for forecasters

Mike Gilliland
Mike Gilliland|January 9, 2012

Have you taken the Myers-Briggs Type Indicator (MBTI) assessment?  It is a psychological test wherefrom you are classified on Extraversion vs. Introversion, Sensing vs. Intuition, Thinking vs. Feeling, and Judging vs. Perceiving.  I, along with roughly 15% of the population, come out an ISTJ or "Guardian Inspector" (the single largest group among the 2^4 = 16 possible classifications).  Apparently this is where all the curmudgeons come from.

What I found most curious was a table of "Famous Examples by Psychological Type" handed out by my examiner.  It placed various historical figures into their Myers-Briggs category (e.g. Benjamin Franklin and Walt Disney were ENTP or "Rational Inventor," and Wolfgang Mozart and Stephen Spielberg were ISFP or "Artisan Composer").  I thought this was more than a little strange, since Benjamin Franklin and Wolfgang Mozart died long before the Myers-Briggs assessment was invented.

I became further suspicious by several of the other classifications of famous people, notably Princess Di, Albert Schweitzer, and Richard Gere all falling under INFP or "Idealist Healer."  Princess Di and Albert Schweitzer I can understand, but Richard Gere a healer??? Perhaps early in his film career (as the "American Gigolo" with a heart of gold), but definitely not since that unfortunate falling out with Fievel.

So is there a Myers-Briggs Type Indicator for people in forecasting? If you've taken the assessment and are willing to share, please submit a comment with your MBTI.

Free 1-on-1 Consulting at IBF Supply Chain Forecasting Conference

Last time on The BFD (the blog) I mentioned a 1/2 day workshop "What Management Must Know About Forecasting" at the upcoming IBF conference (February 26-28 in Scottsdale, AZ), and the complimentary copy of The BFD (the book).  But wait, there's more:

  • Register by January 27 and enjoy early bird pricing and free participation in the IBF Golf Outing.  (Based on past scores from the outing, actually being able to golf is not a requirement.)
  • Round Robin Roundtable Discussions on various topics including Worst Practices in Forecasting.
  • Schedule a 30-minute consultation with me or one of my SAS colleagues from professional services and R&D. We'll be ready to discuss your biggest challenges related to forecasting process, statistical modeling, or whatever you have to throw at us.  Bring us your questions, bring us your data, and if you are already a SAS forecasting customer, go ahead and bring us your code and we'll be happy to take a look.  Send me an email (mike.gilliland@sas.com) to reserve a time, or sign up at the SAS booth at the event.

 

Post a Comment

Institute of Business Forecasting: Workshop and book signing

Mike Gilliland
Mike Gilliland|December 23, 2011

Is there anything you'd like to tell your management?  Of course, in the spirit of the holiday season, I mean is there anything you'd like to tell your management that isn't anatomically impossible?

If so, please join me and Ryan Rickard, Sr. Supply Chain Manager at Newell Rubbermaid, for our 1/2 day workshop on "What Management Must Know About Forecasting." We'll give you plenty of things to talk about.

Ryan and I will be presenting at the Institute of Business Forecasting's Supply Chain Forecasting Conference, being held February 26-28, 2012, in Scottsdale, AZ.  (Register by January 27 and participate in the free golf outing on Sunday February 26.)

The workshop has three sections:

  • Review of fundamental forecasting issues like definition of demand, forecastability, demand volatility, accuracy expectations, and evaluating performance.
  • Step-by-step instructions for conducting Forecast Value Added analysis, including data collection, data analysis, and reporting.
  • Case study on how Newell Rubbermaid is applying these approaches.

While at the event, be sure to visit my colleagues Ed Katz and Phil Weiss at the SAS exhibit booth, and catch Charlie Chase's presentation (with Arnaud Joliff of Nestle) on "Demand Sensing, Shaping and Managing a Promotion-Driven Business: A Case Study from Nestle."

Free Takeaway: The BFD

During the breakfast hour on Tuesday February 28, I'll be handing out signed copies of The BFD (the book) -- the perfect belated Valentine's Day gift for someone you love -- yourself.

 

 

 

 

Post a Comment

My Offering: Forecast Accuracy Objectives for 2012

Mike Gilliland
Mike Gilliland|December 16, 2011

Managing expectations for forecast accuracy is very important, as often those expectations are extreme after management invests in a new system. Software vendors have also been known to make overly (choose one: optimistic? sanguine? idyllic?) accuracy claims as part of their sales pitch.

Of course, there is no arbitrary level of accuracy that can be promised, no matter what management wants or needs. The best way to set expectations for what is reasonable is to run a simple model (e.g. moving average) against your historical data, and see how accurate such a model would have been.

Management has every right to expect a sophisticated new statistical software package to forecast at least this well (it would have been an unfortunate investment if it couldn't!).  However, the accuracy improvement due to a sophisticated model is sometimes surprisingly small -- because sometimes a simple model happens to be the most appropriate for forecasting.

Forecast accuracy is more a function of the “forecastability” of what you are forecasting than of the particular model you use to forecast it. If something is well behaved and easy to forecast, any appropriate model should do reasonably well. If something is highly erratic and difficult to forecast, no model is going to do very well, and you just have to plan your operations around that uncertainty.

Also, forecastability of a product can change over time…you may switch something from stable pricing, to making it highly promoted with frequent price changes. You can probably get more accurate forecasts when pricing is stable (and customer demand is more stable), than when you increase demand volatility by promoting it. So your accuracy expectations would change if there is such a change in how you sell/promote the product.

For setting performance expectations/objectives for the future, they can be as simple as: Do no worse than a moving average. Today I don’t know how accurate a moving average will be in 2012 (so I can’t name a specific accuracy target for next year). However, it is reasonable to expect my forecasting software and forecasting process to NOT make the forecast any worse than that.

Find Three Wrong Ways to Set Accuracy Objectives on the IBF Blog

For more discussion on this topic, including three wrong ways to set accuracy objectives, see my guest post "2012 Forecasting Performance Objectives" on the Institute of Business Forecasting blog.  I want to thank IBF for the special surprise of having my favorite theoretical physicist Stephen Hawking read the blog to us by clicking on the "Listen Now" button near the top of the page.  Check it out!!

Post a Comment

SCM Focus on forecastability and over fitting

Mike Gilliland
Mike Gilliland|December 12, 2011

My Google Alert on "forecastability" paid off with a gem this weekend, in the blog post "Forecastability and Over Fitting" by Shaun Snapp on SCM Focus.  I was not previously familiar with Shaun or this site, but found a lot to like -- in content and attitude.

In his post, Shaun kindly cites the coin-flipping example from The Business Forecasting Deal (the book), which I used to illustrate the nonsense of fitting a model to the random pattern of Heads and Tails in order to forecast future flips.  Says Shaun,

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.  Complicated math is in essence the new mysticism.

Shaun ends with another nice potshot, over a photo of Manhattan,

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.

Please take a few moments away from your busy online shopping this week, and check out Shaun's post.

Post a Comment

High on complexity

Mike Gilliland
Mike Gilliland|November 18, 2011

Paul Goodwin's Hot New Research column is a must-read in each issue of Foresight. The current column, "High on Complexity, Low on Evidence: Are Advanced Forecasting Methods Always as Good as They Seem?" ends with this sage advice:

If the name of a method contains more words than the number of observations that were used to test it, then it's wise to put any plans to adopt the method on hold.

In professional forecasting circles, it has long been recognized that fancier methods are no guarantee of better forecasts (even though fancier methods can often provide a better fit to history!).  As Goodwin notes, "...an improved fit to past data may be associated with poorer forecasts because the method is falsely seeing systematic patterns in the random movements in past data, and assuming that these will continue into the future."

The main point of Goodwin's column is that forecasting researchers -- those publishing articles touting advanced new methods -- should provide reliable evidence that the methods they advocate will actually work. The accuracy of the new method should be compared to the accuracy of appropriate benchmarks (typically simpler methods), to see if the extra complexity of the new method is justified.

Goodwin cites Len Tashman's 2000 article ("Out-of-sample tests of forecast accuracy: An analysis and review," International Journal of Forecasting, 16, 437-450), arguing that forecasting methods need to be tested on a sufficient number of out-of-sample observations. He points to several recently published examples "where grandiose claims of forecasting performance rest on scant evidence, where "big conclusions" are drawn from "small amounts of data."

New forecasting ideas are great -- the bigger and crazier the better. But any new forecasting method should be evaluated (just like we would a new drug or therapy) starting with a null hypothesis:

H0: The new method has no effect on forecasting performance

Without sufficient evidence to reject the null hypotheses, implementing the method might just be wasting our time.

 

Post a Comment

  • About this blog

    Michael Gilliland is a longtime business forecasting practitioner and currently Product Marketing Manager for SAS Forecasting. He initiated The Business Forecasting Deal to help expose the seamy underbelly of the forecasting practice, and to provide practical solutions to its most vexing problems.

  • Subscribe to this blog

    Enter your email address:

    Other subscription options

  • Archives