Memo Take Home

Date: 10-5-2015
To: Dr. Stephen Allen
From: Mukhamed Rashidov
Re: Take Home –Forecasting Methods
First we looked at 3 and 5 periods Moving Averages and found the MAD to be
43.33 and 44.3 respectively which is pretty high taking into consideration the max
(190) and min (20) values for sales (response variable). The forecasts for both
methods used are shown in the Table below. Weighted Moving Average with the
same periods resulted in even higher MAD which disproved the effectiveness of
WMA method in this case (refer to the Table below for results) because the MAD was
even higher than MAD found with moving average. Next, we examined whether
exponential smoothing is suitable or not, the results revealed that neither
exponential smoothing (with alphas = .1 which puts emphasis on more recent data
and .9 which has more weight on old data) nor exponential smoothing with trend
adjustment (beta =.1) are good models to use in this case even though MAD for ES
and ES/TA methods turned out to be 38.2 and 38.25 respectively which is lower
compared to other methods so far. We conducted sensitivity analysis in order to find
appropriate betas and alphas; beta of .1 and alpha of .1 showed the lowest MAD
relative

to


other

levels

of

betas

and

alphas.

Trend projection and TP with Season Adjustment resulted in MAD of 36.1 and
34.9 which is lower compared to the previous methods; forecasts for period 49 were
91.32

and

87.26


respectively.

Simple regression analysis using QWE index as predictor of sales showed R-square
of about 94% which means that the model fits data well and is a good model to
consider; prediction of period 49 was 193.978 and the MAD 7.9 which is much
better compared to all the other previous methods used. Comparatively the CLR
using RTY index showed less promising output as the R-square turned out to be only
9.6%. We also checked RTY index for leading indicators and found that it has a two
period leading indicator that produced output with R-square of 99.85% and
correlation coefficient of 99.9%. MAD in this case was calculated to be only
1.03 which is the lowest compared to all the other methods. We forecasted period
49 sales to be 197.68 which is our best prediction.