In conventional least square regression method, which was
used in SARIMA-MLR model, it has been recognized that the resulting estimates of different
covariates on the conditional mean of sales were not reflecting the real size and nature of these
cov- ariate effects on the lower and upper tail of the sales distribution. In Appendix and Fig.
6, the intercept (c) is different from zero in MLR and is significant from 0.05 to 0.3
quantiles and for 0.95 quantile in QR. The predicted sales of banana (using SARIMA) is
significant in all quantiles. It clearly has a positive impact on the expected sales of banana.
The estimated conditional quantile function reveals that the effect of the covariate is getting
more extreme when the quantiles are higher. The effect on conditional mean and conditional
median are similar. In case of Christmas effect, the upper quantile estimates are
significant and have a positive effect on the sales of banana. For 0.05 quantile, Christmas has a
negative coefficient. This may be due to stock-outs. However, the ordinary least square (OLS)
estimate suggests that the effect of Christmas variable is not different from zero. This is
a salient example of how misleading the ordinary least square estimates can be. Easter effect
variable has an increasing positive effect from lower quantiles to upper quantiles. It is almost
significant for all quantiles except 0.5 quantile. The OLS regression strongly over- estimates
the effect of Easter variable than the median quantile estimate. The school vacation estimate is
not different from zero for any quantiles except 0.3 and 0.05 quantiles and has negative
effect on the sales. The relationship between the sales and before
holiday variable is positive. The effect of before holiday is sig-
nificant for all quantiles and increases from lower quantiles to upper quantiles. The Austrian
holiday variable has a negative effect and is significant from 0.2 to 0.8 quantiles. It has
a decreasing effect from lower quantiles to upper quantiles. The after holiday variable is not
different from zero in most of the quantiles except
0.05, 0.4, and 0.8 quantiles.
According to OLS regression, only the month covariates such as May, June and December are
significant. The OLS estimate for other months are not different from zero. The February month
variable is significant only for 0.3 quantile and has a negative coefficient. The March month
variable is significant only in middle quantiles and has a positive effect. For the March
variable, the coefficients of conditional mean is similar to the coefficient of conditional median of
the sales distribution. The April month variable is significant from 0.7 to 0.9 quantiles and
has a positive effect. The May month variable is significant for all quantiles except
0.95 quantile and has a positive effect. The OLS regression overestimates the effect of this
variable than the median quantile estimate. The June month variable is significant in most of
the quantiles except 0.1, 0.3, 0.8, and 0.95 quantiles and has a positive effect. The July month
is significant only for 0.2 quantile and has a positive effect. The August month variable is
not different from zero, both in OLS and QR. The September month variable is sig- nificant only
for 0.95 quantile and has a negative effect. The October month variable is significant
from 0.1 to 0.5 quantiles and has a negative effect. The November month variable is significant
only for 0.3 quantile and has a negative effect. The December month variable is significant
from 0.1 to 0.7 quantiles and has a negative effect. The discounted sales variable is
significant for lower (0.1–0.3) and upper (0.6–0.9) quantiles, and has a positive effect. The
promotion variable is highly significant and has a positive relationship with the sales of
banana for all quantiles. The estimates for promotion increases from lower quantiles to upper
quantiles. The precipitation covariate is not different from zero in OLS regression. In QR, it
has a significant negative estimate for
0.9 quantile. The sunshine duration variable is significant from
0.05 to 0.95 quantiles and has a negative effect. The fresh snow depth has significant
negative estimates for 0.05, 0.2, and
0.6 quantiles. Conversely, the fresh snow depth is not significant in
OLS regression. The snow depth is significant from 0.3 to
0.8 quantiles and for 0.95 quantile. The estimates of snow depth variables in QR as well as
OLS regression are positive. Theoreti- cally, snow depth has both positive and negative impact
on retail food sales. Agnew and Thornes (1995) also suggested that the adverse weather
usually have negative effect on the larger
In conventional least square regression method, which wasused in SARIMA-MLR model, it has been recognized that the resulting estimates of different covariates on the conditional mean of sales were not reflecting the real size and nature of these cov- ariate effects on the lower and upper tail of the sales distribution. In Appendix and Fig. 6, the intercept (c) is different from zero in MLR and is significant from 0.05 to 0.3 quantiles and for 0.95 quantile in QR. The predicted sales of banana (using SARIMA) is significant in all quantiles. It clearly has a positive impact on the expected sales of banana. The estimated conditional quantile function reveals that the effect of the covariate is getting more extreme when the quantiles are higher. The effect on conditional mean and conditional median are similar. In case of Christmas effect, the upper quantile estimates are significant and have a positive effect on the sales of banana. For 0.05 quantile, Christmas has a negative coefficient. This may be due to stock-outs. However, the ordinary least square (OLS) estimate suggests that the effect of Christmas variable is not different from zero. This is a salient example of how misleading the ordinary least square estimates can be. Easter effect variable has an increasing positive effect from lower quantiles to upper quantiles. It is almost significant for all quantiles except 0.5 quantile. The OLS regression strongly over- estimates the effect of Easter variable than the median quantile estimate. The school vacation estimate is not different from zero for any quantiles except 0.3 and 0.05 quantiles and has negative effect on the sales. The relationship between the sales and beforeholiday variable is positive. The effect of before holiday is sig-nificant for all quantiles and increases from lower quantiles to upper quantiles. The Austrian holiday variable has a negative effect and is significant from 0.2 to 0.8 quantiles. It has a decreasing effect from lower quantiles to upper quantiles. The after holiday variable is not different from zero in most of the quantiles except0.05, 0.4, and 0.8 quantiles.According to OLS regression, only the month covariates such as May, June and December are significant. The OLS estimate for other months are not different from zero. The February month variable is significant only for 0.3 quantile and has a negative coefficient. The March month variable is significant only in middle quantiles and has a positive effect. For the March variable, the coefficients of conditional mean is similar to the coefficient of conditional median of the sales distribution. The April month variable is significant from 0.7 to 0.9 quantiles and has a positive effect. The May month variable is significant for all quantiles except 0.95 quantile and has a positive effect. The OLS regression overestimates the effect of this variable than the median quantile estimate. The June month variable is significant in most of the quantiles except 0.1, 0.3, 0.8, and 0.95 quantiles and has a positive effect. The July month is significant only for 0.2 quantile and has a positive effect. The August month variable is not different from zero, both in OLS and QR. The September month variable is sig- nificant only for 0.95 quantile and has a negative effect. The October month variable is significant from 0.1 to 0.5 quantiles and has a negative effect. The November month variable is significant only for 0.3 quantile and has a negative effect. The December month variable is significant from 0.1 to 0.7 quantiles and has a negative effect. The discounted sales variable is significant for lower (0.1–0.3) and upper (0.6–0.9) quantiles, and has a positive effect. The promotion variable is highly significant and has a positive relationship with the sales of banana for all quantiles. The estimates for promotion increases from lower quantiles to upper quantiles. The precipitation covariate is not different from zero in OLS regression. In QR, it has a significant negative estimate for0.9 quantile. The sunshine duration variable is significant from0.05 to 0.95 quantiles and has a negative effect. The fresh snow depth has significant negative estimates for 0.05, 0.2, and0.6 quantiles. Conversely, the fresh snow depth is not significant inOLS regression. The snow depth is significant from 0.3 to0.8 quantiles and for 0.95 quantile. The estimates of snow depth variables in QR as well as OLS regression are positive. Theoreti- cally, snow depth has both positive and negative impact on retail food sales. Agnew and Thornes (1995) also suggested that the adverse weather usually have negative effect on the larger
การแปล กรุณารอสักครู่..
