Non Chosen forecasting methods, alternative to the Flostock approach

These notes have been published in Flostock News, the monthly newsletter by Flostock, about alternative forecasting methods.

 

 

1 Extrapolation

Flostock does not extrapolate straight.

Simplest forecast is a quantitative extrapolation of the past. This assumes that the future is somehow a logical function of the past. The simplest extrapolation is a straight line, but it can also be a bended curve or even an S-curve. We can still call it extrapolation. One may add trends, seasonality and cyclical variations. Major advantage of extrapolation is that it is simple and easy to understand. The disadvantage is not that Black swans are not included, because that is true for all forecasting systems. Nay, the major disadvantage is that all non-linear effects are omitted. For example the inventory effects of the supply chains are certainly non-linear and thus can be surprisingly counter-intuitive. 

The simple extrapolation has been made more complex over time by using a weighted moving average,  by Exponential smoothing, by Autoregressive moving average (ARMA), or by Autoregressive integrated moving average (ARIMA). It all remains the same: extrapolation of the past. 

2 Customer collaboration

Hé, collaborator!

It is possible to obtain a qualitative view of the future by asking your customers what they expect, either direct or via a sort of market research. When you ask enough customers you get an average view that is one step closer to the end market, so it is probably better than sailing on your own feeling alone. Problem with this is that customers, like yourself, don’t know what the future will bring. They are as susceptible to emotions, bias, prejudices and scares projected by the media as you are. Customer collaboration is close to the purchase managers index (PMI): it is a sentimental forecast.  And like simple extrapolation: customer sentiment also forecasts in a linear way and sentiment never includes supply chain effects. 

3 Bottom Up

Bottoms up.

Many companies enter expected sales on product, customer or even product-customer level into their ERP system. Some companies do this up to 18 months into the future. When the whole crew of sales managers is monthly filling up a database, bottom up, it seems that the combined level must be reliable because it is based on very careful thinking by people who know their customers best, a bit like crowdsourcing. Problem is that the sales managers do not know what their customers will order, because their customers don’t know either. And the people involved often adapt their detailed forecast to the expected total, then split it up until it fits. Another problem is that it is a lot of work, and it gives the false impression that it is accurate and well supported by the team. And it is only linear, and it does not take the supply chain effects into account. Main problem is that it is sentimental. Advantages of this system? Don’t know. It sounds solid because it is expensive. I’d say that in most cases this can only be used to break up a forecast on total level into customer/product combinations for production planning. Don’t you ever base your capacity expansions on this. I expect that many companies use this because everybody is doing it, it is massive, expensive and sounds as if everything is under control. “Nobody ever got fired for properly implementing a huge and expensive ERP system. “ Or were they?

4 Point of sale

Point-of-Sale

Some retailers are prepared to collect point-of-sale info and make that available to their suppliers. The biggest advantage of this is that it is accurate and –when delivered in time- gives good steering info. Main disadvantage is that it is not a forecast. It gives the supplier info about what has happened in the –hopefully recent-  past. Due to info delays in the chain, fresh POS info can still arrive before the replenishment order, but only marginally. For companies more upstream in the chain who produce half-products that are assembled by their customers into finished products that are sold in retailer shops, the point-of-sale info cannot be obtained because the half-products are used in a large variety of products and sold in many different shops. So Point-of-sale is useful only in certain situations and mainly interesting for an OEM, who can relate the info to his products. 

5 Life cycle Analogy

Whale Curve

Life cycle analogy (LCA) is a qualitative forecasting method based on the idea that similar products will have a similar life cycle pattern, so if you know how a previous product did in the past, you have a fair assumption how a new product will do. The main disadvantage is immediately clear: this is a very rough approximation. Results from the past never give a guarantee for the future. Times change. Preferences change. Competing technologies change. In reality a lot of people unconsciously use LCA and call it “experience”. What can be valuable however is the shape of the product life cycle. This shape can be used in combination with other, more quantitative methods, as a sort of trend line representing Market Share development and the penetration of a product into an end market. The Flostock models can work with this.  

6 Naïve forecasting

Predict no evil, see no evil

The naïve forecasting method is also the simplest. In effect it means that you assume that tomorrow is today. Forecast by doing nothing. Thirty years ago the best weather forecast was by looking out of the window. If one said that tomorrow would give the same weather as today, your accuracy was higher than that of the weather channel.  Advantages are of course that it is cheap, fast and simple. And it does not give a false sense of sophistication and wisdom. It forces the organization to be flexible and adaptive. And in some situations it can work, e.g. if all alternatives fail, or if a company is monopolist, or so rich it does not matter, or if the business is very stable. That means that for most companies in the real world it is not good enough. Call Flostock if you want to change.  

7 Pre-cursors aka Promotions, innovation and product introductions

Product launch.

This seventh alternative to Flostock modeling is used especially in companies that have a lot of new product introductions.  To estimate the success of a new product, companies consider what happened last time this kind of product was introduced, and adjust the estimate for any plausible differences. This can be based on their own product line or on product introductions from other companies. The obvious disadvantage is that success in the past does not guarantee the future, because circumstances differ and –hopefully-  certainly for new product. Also competition is evolving, customer preferences change, technological progress does not stop, etc. So all in all this forecasting method is pretty limited in applicability and usefulness.  

8 Leading Indicators

Leading into the bush

Managers would like to find an indicator that is correlated to their business and  earlier in time: a so-called leading indicator. There are several issues with this approach:  first issue is that the pattern of the available indicator is often completely  different from the business, even if the indicator represents the correct driving force behind the business. This can be due to delays, hold-ups and feedback loops or because of an interaction with other driving forces, other indicators. A second issue is the pseudo-leading indicator:  assume variable B is the growth of variable A. In a graph the variable B will seem to lead A, but in reality it is the opposite.  Please try this at home with any curve. It is optically misleading. A third issue is that the correlation is based on the past: any new event or development is missing.  E.g. the Lehman Wave was not part of any indicator during the last 10 years.  If you want to see the causal relation between your indicators and your business, taking delays, hold-ups, and feedback loops into account, call Flostock!

9 Consensus Forecasting

Delphi method

This method is about getting a higher quality forecast by averaging the forecasts from different experts, made with different methods. Big advantage is that some biases and systematic errors will be averaged out; in that sense it is similar to crowdsourcing.  One example is the Delphi method, in which there are several rounds, with result sharing after each round, so the group works towards consensus.  An obvious problem with such open method is that the loudest voices will dominate, so it will not avoid herd behavior.  Another, more basic problem is that it is sentiment based, which makes it ‘linear’ by definition, thus overshooting at each ‘turn of the road’. This is because all variants of sentimental forecasting miss the non-intuitive effects of the Stocks as discussed in Law 19

10 Judgment

Impartial judgment

Judgment is one of the so-called Qualitative Methods and is used a lot because it is so simple. It means that the forecast is based on the opinions of (presumed) experts, such as customers, industry leaders, experienced sales people, or simply ‘the boss’. Choice of which opinion to believe is subjective and the method is sentimental and opportunistic. The organizer may even be ‘leading the witness’. In addition, many experts have imperfect info and are simply guessing. This method may also include the infamous ‘Analyst Override’ or ‘Executive Approval’, so introducing bias and wishful thinking. Finally, as with all alternative methods, it can only forecast linear.