3
$\begingroup$

I came across the Lewandowski method of demand forecasting in JDA Demand. Please help me understand at a high level the methodology it uses. I found a paper by Robert Hyndman titled "A state space framework for automatic forecasting using exponential smoothing methods" and it uses this method as one of methods they compare their algorithm to in the paper. Currently for us this is a black box, we want to get some high level understanding so that we can better fine tune the parameters they have provided as part of the software. It would be great if you can share some thoughts about the Lewandowski algorithm and point to some references that I could use for further research.

$\endgroup$
  • $\begingroup$ There is no published or described process of how the Lewandowski method works. It is a mystery. $\endgroup$ – Tom Reilly Mar 12 '13 at 12:53
  • $\begingroup$ Refer to this presentation for Lewandowski alogorithm. youtube.com/watch?v=qKt2P1De6U8 $\endgroup$ – user37658 Jan 22 '14 at 10:51
2
$\begingroup$

I worked with the Lewandowski algorithim on the JDA plattform for years, it's been a while but I will try and explain.

In Lewandowski you have a forecasted mean value (history sensitivity, trend,...), a seasonal pattern that can be influenced by history, two types of events (one absolute and one dynamic) and causal factors. @IrishStat, yes it is similar to Holt-Winters.

Lewandowski gives the user alot of room to influence the forecast, both on perfecting the statistical forecast, but also in creating good seasonal patterns. For better and worse...it was easy for our users to loose their way with all the parameters.

I still have the manual, but I get get hold of it at the moment.

$\endgroup$
2
$\begingroup$

As I know from Lewandowsky Algorithm, it works like Holt-Winters algorithm. You will define three parameters:

  1. α is the data smoothing factor and it's 0 < α < 1
  2. β is the trend smoothing factor, 0 < β < 1,
  3. γ is the seasonal change smoothing factor, 0 < γ < 1.

If you select the big number (near to 1) for α, it means you rely more on recent past data rather than old past data.
If you choose the big number (near to 1) for β, it means you rely more on past data's trend and you believe the trend will go on in future too. (you increase the weight of trend smoothing)
If you choose the big number (near to 1) for γ, it means you rely more on past data's seasonality and you believe the seasonality factors will remain in future.
My suggestion is to start with some numbers for α,β and γ then after each period try to calculate your error, and find the numbers which reduce your errors.
I used this model in Health-care and it gives me the accurate numbers.

$\endgroup$
2
$\begingroup$

Like all the gentlemen mentioned here, Lewandowski is very similar to the Holt-Winters. The main difference is that Holt-Winters is a static algorithm in that the Alpha, Beta, and Gamma values do not change until they are manually altered.

The Lewandowski, on the other hand, is an adaptive algorithm where the Alpha, Beta, and Gamma are tweaked automatically by proprietary algorithms. Hope that helps!

Jimmy Varghese, works for JDA Software

$\endgroup$

Your Answer

By clicking “Post Your Answer”, you agree to our terms of service, privacy policy and cookie policy

Not the answer you're looking for? Browse other questions tagged or ask your own question.