You are on page 1of 31

Causal Methods

1 Assumes that demand forecast is highly correlated with certain factors


(Eg: Economy, interest rates, Inflation, Unemployment, Demograhics, Promotio
2 Factors all relevant environmental factors which may impact the dema

Simulation Forecasting Methods


These methods imitate consumer choices that give rise to demand to ar
Eg: What will be the impact of price promotions?
Eg: What will be the impact of Competitor Opening a store near by?
Combines Time-Series and Causal Methods
FACTORS TO BE CONTEMPLATED
1 HISTORICAL DEMAND DATA
2 ADVERTISING & MARKETING EFFORTS
3 STATE OF THE ECONOMY
4 PLANNED PRICING
5 COMPETITORS ACTIONS
with certain factors in the environment
mograhics, Promotions, Competition)
y impact the demand

rise to demand to arrive at a forecast


Quarterly Demand for Dell Inspiron 15 Laptops in Market Segment (India South) through Reta
Year Quarter Period t
0
2017 1 1
2017 2 2
2017 3 3
2017 4 4
2018 1 5
2018 2 6
2018 3 7
2018 4 8
2019 1 9
2019 2 10
2019 3 11
2019 4 12
2020 1 13
2020 2 14
2020 3 15
2020 4 16
on 15 Laptops in Market Segment (India South) through Retail
Demand Dt Year
Quarter
2017
8000 1 8000
13000 2 13000
23000 3 23000
34000 4 34000
10000
18000
23000 1
38000 2
12000
13000 Appropriate Quantitative Method for Sa
32000
41000
?
?
?
?
Year
2018 2019 2020
10000 12000
18000 13000
23000 32000
38000 41000

Presence of Trend? Yes/No


Presence of Seasonality? Yes/No

ate Quantitative Method for Sales Forecasting?


Method 1:
Moving Average
( Use When no observable Trend or Seasonality

Systematic Component of Demand = Level


In this method, we estimate the level in period t as the average demand over
This represent N-period moving average.

Lt = ( Dt + Dt-1 + ………. + D t-N+1)/N

Current forecast for all future periods is the same and is based on the c
Ft +1 = Lt Ft +n = Lt
After observing Demand for period t +1, we revise the estimate as below:

Lt +1 = ( Dt +1 + Dt + ………. + D t-N+2)/N
Eg: 4-period Moving average for Natural Gas.com

L4 = (D4 + D3 + D2 + D1) /4 = (34000 + 23000 +13000 +8000

F5 = L4 F5 = 19500

E5 = F5 - D5 = 19,500 - 10,000

L5 = (D5 + D4 + D3 + D2)/4 = (10000 + 34000 +23000 +1300

Applying Moving Average Method


Quarterly Demand for Dell Inspiron (India South Segment) No. of Demand Pe

Year Quarter Period t Demand Dt Level Forecast Error


0
2017 1 1 8000
2017 2 2 13000 0
2017 3 3 23000 0
2017 4 4 34000 19500
2018 1 5 10000 20000 19500 9500

2018 2 6 18000 21250 20000 2000


2018 3 7 23000 21250 21250 -1750

2018 4 8 38000 22250 21250 -16750

2019 1 9 12000 22750 22250 10250

2019 2 10 13000 21500 22750 9750

2019 3 11 32000 23750 21500 -10500

2019 4 12 41000 24500 23750 -17250


2020 1 13 24500
2020 2 14 24500
2020 3 15 24500
2020 4 16 24500

Conclusions:

Average error per forecast is 49% % of actual demand.


Average error per forecast is 9719 units of actual demand.
The direction of error is both positive and negative (TS) Highest= 2.2
verage
e Trend or Seasonality)

verage demand over the most recent N periods..

me and is based on the current estimate level

timate as below:

Ft + 2 = Lt + 1

0 + 23000 +13000 +8000)/4 19500

F5 = 19500

0 - 10,000 = 9500

0 + 34000 +23000 +13000)/4 = 20000

age Method
No. of Demand Periods for Moving Average = 4
Cummulative Absolute Cumm Ab
Error Dev Dev MAD % Error MAPE TS

9500 9500 9500 9500 95.0% 95% 1.0


11500 2000 11500 5750 11.1% 53% 2.0
9750 1750 13250 4417 7.6% 38% 2.2
-7000 16750 30000 7500 44.1% 39% -0.9
3250 10250 40250 8050 85.4% 49% 0.4
13000 9750 50000 8333 75.0% 53% 1.6
2500 10500 60500 8643 32.8% 50% 0.3
-14750 17250 77750 9719 42.1% 49% -1.5

Lowest= -1.5
1
Method 2:
Simple Exponential Smooting
( Use When no observable Trend or Seasonality)

Systematic Component of Demand = Level


The initial estimate of level L0, is taken to be the average of all historical data because deman
assumed to have no observable trend or seasonality

L0 = (D1 + D2 + …………… + Dn)/n

Ft +1 = Lt Ft +n = Lt

After observing Demand for period t +1, we revise the estimate as below:

Lt +1 = α Dt +1 + ( 1- α)Lt

Where α is a smoothing constant for the level, 0 < α < 1.


The revised value of the level is weighted average of the observed value D t+ 1 and the o

In the below example

L0 = (8000 + 13000 +23000 + 34000 + 10000 + 18000 + 23000 + 38000 + 12000 + 13000 + 32

L0 = 22083.33 F1 = L0 = 22083

Applying Exponential Smooting Method


Quarterly Demand for Dell Inspiron (India South Segment)

Year Quarter Period t Demand Dt Level Forecast Error


0 22083
2017 1 1 8000 21638 22083 14083
2017 2 2 13000 21365 21638 8638
2017 3 3 23000 21416 21365 -1635
2017 4 4 34000 21814 21416 -12584
2018 1 5 10000 21441 21814 11814
2018 2 6 18000 21332 21441 3441
2018 3 7 23000 21385 21332 -1668
2018 4 8 38000 21910 21385 -16615
2019 1 9 12000 21597 21910 9910
2019 2 10 13000 21325 21597 8597
2019 3 11 32000 21662 21325 -10675
2019 4 12 41000 22274 21662 -19338
2020 1 13
2020 2 14
2020 3 15
2020 4 16

Conclusions:

Average error per forecast is 59% % of actual demand.


Average error per forecast is 9916 units of actual demand.
The direction of error is both positive and negative (TS)
Smooting
end or Seasonality)

data because demand has been

1.760375
Ft + 2 = Lt + 1

lue D t+ 1 and the old estimate Lt

12000 + 13000 + 32000 + 41000)/12

Method
Smoothing Constant for Level = 0.03162
Cummulative Absolute
Err Dev MAD % Error MAPE TS

14083 14083.0 14083 176.0% 176% 1.0


22721 8637.7 11360 66.4% 121% 2.0
21085 1635.3 8119 7.1% 83% 2.6
8502 12583.6 9235 37.0% 72% 0.9
20316 11814.2 9751 118.1% 81% 2.1
23757 3440.7 8699 19.1% 71% 2.7
22089 1668.1 7695 7.3% 62% 2.9
5473 16615.4 8810 43.7% 59% 0.6
15383 9910.0 8932 82.6% 62% 1.7
23980 8596.6 8898 66.1% 62% 2.7
13305 10675.2 9060 33.4% 60% 1.5
-6033 19337.6 9916 47.2% 59% -0.6

% of actual demand.
units of actual demand.
Highest: 2.9 Lowest: -0.6
Method 3:
Trend Correct Exponential Smoothi
( Use When no observable Se

Systematic Component of Demand = Level + Trend


Initial Estimate of Level (L0) =
Run a linear regression between Dt and time period t of the follo

Constant b measures the estimate of demand at period t = 0


It is also the estimate of the initial Level L0
Constant a measures the rate of change in demand period and is the initial e

In period t, given estimates of Level Lt, and trend Tt, the forecast for future periods:

Ft+1 = Lt + Tt Ft+

After observing demand for period t, we revise estimates for level

Lt+1 = α Dt+1 + (1- α) (Lt + Tt)

Tt+1 = β (Lt+1 - Lt) + (1- β)Tt


Where α is Smoothing Constant for Level
Quarterly Demand for Dell Inspiron (India South Segment)
Period (t) Demand (D) Level (L) Trend (T) Forecast (F) Error E(t)

0 12015 1549
1 8000 13564 1549 13564 5564
2 13000 15113 1549 15113 2113
3 23000 16662 1549 16662 -6338
4 34000 18211 1549 18211 -15789
5 10000 19760 1549 19760 9760
6 18000 21309 1549 21309 3309
7 23000 22858 1549 22858 -142
8 38000 24407 1549 24407 -13593
9 12000 25956 1549 25956 13956
10 13000 27505 1549 27505 14505
11 32000 29054 1549 29054 -2946
12 41000 30603 1549 30603 -10397
13
14
15
16

Min TS = Max TS =
Slope
Estimate of change in demand(Trend) at t = 0 denoted by a 1549 Estimate o
Holt's Model Smoothing Constant of Level (α) =
D= at + b

a = rate of change in demand per period = Initial estimate of the trend T 0


b= estimate of demand period (t = 0) = estimate of the initial Level L 0
onential Smoothing [HOLT'S Model)
hen no observable Seasonality)

me period t of the following form:

Dt = at + b

d and is the initial estimate of the T 0

for future periods:

Ft+n = Lt + nTt

estimates for level and trend as:

β is Smoothing Constant for Trend

CUMULATIVE Absolute Error


MAD (t) % Error MAPE (t) Tracking Signal (TS)
ERROR A(t)

5564 5564 5564 70% 70% 1


7677 2113 3839 16% 43% 2
1339 6338 4672 28% 38% 0.286652806652805
-14450 15789 7451 46% 40% -1.93930861880182
-4690 9760 7913 98% 51% -0.59270606256997
-1381 3309 7145 18% 46% -0.19328530651081
-1523 142 6145 1% 39% -0.24789335356437
-15117 13593 7076 36% 39% -2.13629812641548
-1161 13956 7840 116% 48% -0.14805761099366
13344 14505 8507 112% 54% 1.56859252764115
10397 2946 8001 9% 50% 1.29945575168505
0 10397 8201 25% 48% 0

Since Ft+n = Lt + nTt

Intercept

Estimate of demand at t = 0 or estimate of initial Level(L0)= 12015


0 Const for Trend (β)= 0.227
Method 4:
Trend and Seasonality Corrected Exponential Smoothing
( Use When no both Trend and Seasonality is Observ

Systematic Component of Demand = Level + Trend + Seasonality


Initial Estimate of Level (L0) =
Run a linear regression between Dt# and time period t of the following form:

Dt# = at + b
Constant b measures the estimate of demand at period t = 0
It is also the estimate of the initial Level L0
Constant a measures the rate of change in demand period and is the initial estimate of
Eg: DeSeaonalised Demand (D3#) =
In period t, given estimates of Level Lt, trend Tt, and Seasonality Factor St the forecast for future pe

Ft+1 = (Lt + Tt)St +1 Ft+n = (Lt + nTt)St + n

After observing demand for period t, we revise estimates for level and trend as:

Lt+1 = α (Dt+1/St+1) + (1- α) (Lt + Tt)

Tt+1 = β (Lt+1 - Lt) + (1- β)Tt

St+p+1 = γ(Dt+1/Lt+1) + (1- γ)St+1

Quarterly Demand for Dell Inspiron (India South Segment)


Deseasonalized
Period Demand Deseasonalized Demand ( Rev )
Level (L) Trend (T)
(t) (D) Demand (D#)

0
1 8000
2 13000
3 23000 #DIV/0!
4 34000 #DIV/0!
5 10000 #DIV/0!
6 18000 #DIV/0!
7 23000 #DIV/0!
8 38000 #DIV/0!
9 12000 #DIV/0!
10 13000 #DIV/0!
11 32000
12 41000
13
14
15
16

Periodicity =

Estimate of change in demand (Trend) at t = 0 denoted by a =


Estimate of initial Demand or Initial Level L0 (b) =
Winter's α = Smoothing Constant of Level =
Winter's Model

D= at + b
a = rate of change in demand per period = Initial estimate of the trend T0
b= estimate of demand period (t = 0) = estimate of the initial Level L0
ponential Smoothing [WINTER'S Model)
and Seasonality is Observable)

the following form:

is the initial estimate of the T 0


(D1 + D5 + 2(D2 +D3 + D4) )/8
St the forecast for future periods:

+ nTt)St + n

for level and trend as:

Tracking
Seasonal Season Seasonal Forecast Error E(t) Absolute MAD (t) % Error MAPE
Signal
Factor 1 Factor(St) Factor (St) (F) Error A(t) (t)
(TS)
Since, Ft+n = (Lt + nTt)St + n

Min TS = Max TS =

end) at t = 0 denoted by a = #DIV/0! slope


r Initial Level L0 (b) = #DIV/0! intercept
β = Smoothing constant of Tre
γ = Smoothing Constant for Season Factor

mate of the trend T0


initial Level L0
Method 1:
Forecas
( Use When ob

Applying Seasonali
Quarterly Demand for Dell Inspiron (India South Segment)
Avg Seasonality
Year Quarter Period t Demand Dt Seasonal Index of Period Index
0
2017 1 1 8000 0.41 0.45
2017 2 2 13000 0.67 0.67
2017 3 3 23000 1.18 1.17
2017 4 4 34000 1.74 1.71
2018 1 5 10000 0.45 0.45
2018 2 6 18000 0.81 0.67
2018 3 7 23000 1.03 1.17
2018 4 8 38000 1.71 1.71
2019 1 9 12000 0.49 0.45
2019 2 10 13000 0.53 0.67
2019 3 11 32000 1.31 1.17
2019 4 12 41000 1.67 1.71
2020 1 13
2020 2 14
2020 3 15
2020 4 16

Conclusions:

Average error per forecast is 12% % of actual demand.


Average error per forecast is 2024 units of actual demand.
The direction of error is both positive and negative (TS)
Forecasting Seasonality
( Use When observable Seasonality only)

Smoothing constant (alpha) 0.3637181

Applying Seasonality Index Method


No. of Demand Periods for Moving Average =
De-seasonalised Cummulative Absolute
Demand level Forecast Error Error Dev

22094 22094
17784 20526 9938 1938 1938 1938
19439 20131 13727 727 2665 727
19606 19940 23616 616 3281 616
19903 19926 34064 64 3345 64
22231 20765 8964 -1036 2308 1036
26916 23002 13886 -4114 -1805 4114
19606 21767 26984 3984 2178 3984
22244 21940 37184 -816 1362 816
26677 23663 9869 -2131 -768 2131
19439 22127 15825 2825 2057 2825
27278 24000 25957 -6043 -3986 6043

24000 24000 41000 0 -3986 0.006136832

Highest= 4.0 Lowest= -2.0


=
Cumm Ab
Dev MAD % Error MAPE TS

1938 1938.37 24.2% 24% 1.0


2665 1332.72 5.6% 15% 2.0
3281 1093.69 2.7% 11% 3.0
3345 836.18 0.2% 8% 4.0
4381 876.24 10.4% 9% 2.6
8495 1415.80 22.9% 11% -1.3
12478 1782.62 17.3% 12% 1.2
13294 1661.76 2.1% 11% 0.8
15425 1713.85 17.8% 11% -0.4
18249 1824.95 21.7% 12% 1.1
24292 2208.41 18.9% 13% -1.8

24292.47845 2024.37 0.0% 12% -2.0


1

You might also like