LAHORE UNIVERSITY OF MANAGEMENT

Problem Solving and Decision Making PSDM

Assignment: Estimating WAPDA Bill for my House in Nathia Gali

2023-04-0010
10/24/2021

[Type the abstract of the document here. The abstract is typically a short summary of the contents of the document.
Type the abstract of the document here. The abstract is typically a short summary of the contents of the document.]
2023-04-0010

Contents

Assignment: Estimating WAPDA Bill for my House in Nathia Gali...........................................................0

Section 1......................................................................................................................................................2
Section 2:.....................................................................................................................................................3
Replacing Missing Values.........................................................................................................................3
Section 3......................................................................................................................................................4
Forecast of monthly Wapda Bill for 2001 using Regression Analusis.......................................................4
Time Series Pattern:................................................................................................................................5
Trend:......................................................................................................................................................5
Seasonality:.............................................................................................................................................5
Regression Model....................................................................................................................................5
Regression Equation................................................................................................................................5
Interpretation..........................................................................................................................................5
Forecast Through Regression..................................................................................................................6
Section 4:.....................................................................................................................................................6
Time Series Decomposition:....................................................................................................................6
Regression Model....................................................................................................................................8
Regression Equation................................................................................................................................8
Interpretation..........................................................................................................................................8
Forecast Through Classical Decomposition.............................................................................................8
SECTION 5....................................................................................................................................................9
Modles Comparison.................................................................................................................................9
Section 6....................................................................................................................................................10
Logistic Regression................................................................................................................................10
Regression Model..................................................................................................................................10
Regression Equation..............................................................................................................................11
Interpretation........................................................................................................................................11

1|Page
2023-04-0010

Section 1
This case is about electricity consumption of a house located near Nathia Gali. We have 10 years previous data
of different variable related to Wapda electricity bill.

In this case below Forecasting will be done;

What How
Inserting Estimated It will be done through
Missing Values of Wapda
Bill Data Points
Forecasted values of It will be done through Multiple Regression using Dummy Variable
monthly Wapda bill of of Months, as to explain Seasonality. And Time Period for Trend.
2001
Seasonal Indices based on I will use Classical Decomposition Method for Forecasting of
10-year data. And monthly Monthly Wapda bills of 2011. Seasonal Indices calculated in
Wapda Bill forecast of method will be explained.
2001
Regression and Classical Forecasts from both methods will be compared for accuracy and
Decomposition Forecasts explanation of Trend/Seasonality.
of 2001 Wapda bill
comparison.

Probability of 2000 KwH Probability will be estimated using Logistic Regression. All
and 4000KwH electricity elements of models including Odds ratios will be explained.
usage by less than 5
family members.

2|Page
2023-04-0010

Section 2:
Replacing Missing Values
Missing values can be replaced through number of forecasting methods.

1. By taking Mean Value of preceding and successive values

2. Regression Analysis of data
3. Classical Decomposition method in case of Time Series Data.

We will use Mean value for replacing missing values, as we have 120 data point and a percentage error will not
affect our following Forecasts significantly.

Years Months Monthly Bill

1,993 Dec 10,597.50
1,994 Jan 12811.75
1,994 Feb 15,026.00
1,994 Mar 17,872.25 Missing Values
1,998 Apr 4,548.75
1,998 May 2988.75
1,998 Jun 1,428.75

Section 3
Forecast of monthly Wapda Bill for 2001 using Regression Analysis
First, we will have to analyze our electricity data, for determination of regression approach and best possible
model development. Below is Graphical representation of our available Data.

50,000
45,000
40,000
35,000
30,000
25,000 No.
20,000 Monthly Bill
15,000
10,000
5,000
0
1 8 15 22 29 36 43 50 57 64 71 78 85 92 99 106 113 120

FIGURE 01: MONTHLY WAPDS BILL TIME SERIES PLOT

3|Page
2023-04-0010

Time Series Pattern:

Visual representation of data is showing following attributes.

Monthly Bill is having Time Series Pattern with both Trend and Seasonality.

Trend:
Data is showing gradual upward shift till 2007, then may due to replacement of significantly more energy
efficient heat Pump in 2007 and other in 2008 level of trend is changing.

Seasonality:
WAPDA bill reading are having a repeating pattern over a one-year period, showing yearly Seasonal attribute.

Regression Model
We must use Multiple Regression for Forecasting of 2001 monthly bills as per instructions. Months will be used
as Dummy variables as to handle Seasonality and Time variable will be used to predict Trend present in Data.

Model Summary
Adjusted R Std. Error of
Model R R Square Square the Estimate
1 .627a .393 .325 6457.285
a. Predictors: (Constant), snum, jul, jun, aug, may, sep, apr,
oct, mar, feb, nov, dec

FIGURE 2: MULTIPLE REGRESSION MODEL SUMMARY

Regression Equation
Y = b0 + b1F + b2M + b3A + b4M + b5J + b6J + b7A + b8S + b9O + b10N + b11D + b12*t

Interpretation
R Square: 39% variability of Dependent Variable is explained by selected model.

Monthly Dummy Variables: With 95% certainty, amount of Monthly WAPDA Bill will be “bn” amount higher
in respective month than our base case (January), If “t” is kept constant.

4|Page
2023-04-0010

Time Variable: With 95% certainty, value of Monthly WAPDA Bill will increase “b12” amount higher with
every unit increase in time.

For details of model please refer to Exhibit 1 and Exhibit 2.

Forecast Through Regression

Months Forecasted Bill Through Regression
Jan 14,805.71
Feb 14,796.38
Mar 14,787.05
Apr 14,777.73
May 7,797.30
Jun 4,185.23
Jul 5,467.01
Aug 5,444.68
Sep 4,716.16
Oct 4,193.23
Nov 4,034.03
Dec 14,703.10
FIGURE 3: FORECASTED 2001 MONTHLY BILL THROUGH REGRESSION

*In months having insignificant coefficients have been replaced with base case, as insignificant coefficients
imply that value is not significantly different than base case. Slight change in insignificant monthly bills values
are explained by “t”.

Section 4:
Develop and explain seasonal indices for the electricity usage. Use classical decomposition to forecast the
WAPDA Bill for all of 2001.

Time Series Decomposition:

We will use multiplicative decomposition model as it can explain level of fluctuation that are depending on
time, as shown in FIGURE 1.

Yt = Trendt x Seasonalt x Irregulatt

5|Page
2023-04-0010

We will follow below steps for calculating Seasonal Indices and Forecasting 2001 Bills

 12 months moving average of bills will be taken

 Central moving average will be calculated
 Seasonal-Irregular values will be calculated by dividing Bills from Central Moving average.
 Then Seasonal Indices will be calculated by taking mean of all Seasonal-Irregulars of corresponding
months.
Months 1991 1992 1993 1994 1995 1996 1997 1998 1999 2000 Mean Adjustment Factor Adjusted Seasonal Indices
Jan 1.54543 1.66003 1.61588 1.71429 1.76501 1.64614 1.81469 1.86496 1.36862 1.66612 1.007986 1.68
Feb 1.9434 2.05224 1.69398 2.04034 2.10823 2.3329 1.91946 3.89967 1.66069 2.18343 1.007986 2.20
Mar 1.3551 1.92331 2.04335 1.97717 1.95059 1.10587 1.00132 0.98168 1.1775 1.50177 1.007986 1.51
Apr 1.22226 1.25849 1.19368 1.10474 1.65811 0.76245 0.94824 0.80416 2.07112 1.22481 1.007986 1.23
May 1.04003 0.66724 0.71987 0.8451 0.82828 0.70083 2.21115 0.21396 1.18242 0.93432 1.007986 0.94
Jun 0.51953 0.43258 0.54773 0.51092 0.51988 0.50984 0.32114 0.06406 0.72769 0.46149 1.007986 0.47
Jul 0.69611 0.52243 1.05594 0.74708 0.69813 0.46429 0.46381 0.4588 0.15103 0.58418 1.007986 0.59
Aug 0.75137 0.61372 0.72026 0.56799 0.71213 0.64186 0.66248 0.58141 0.61 0.65125 1.007986 0.66
Sep 0.43376 0.43878 0.63655 0.41973 0.74445 0.39272 0.62744 0.51196 0.52 0.52504 1.007986 0.53
Oct 0.68256 0.48345 0.37876 0.49467 0.41234 0.41372 0.49344 0.54007 1.03961 0.54873 1.007986 0.55
Nov 0.50064 0.47375 0.4707 0.47711 0.3674 0.5691 0.55088 0.39504 0.8001 0.51164 1.007986 0.52
Dec 1.54633 1.04398 1.15251 0.70791 0.95016 1.8462 1.17117 0.69634 0.89484 1.11216 1.007986 1.12
0.99208 1.008742395 1.00

Total of Means 11.9049

Adjustment Factor 1.00799

FIGURE 4: SEASONAL INDICES CALCULATION

Adjusted Seasonal Indices

2.50
2.00
1.50 Adjusted Seasonal Indices

1.00
0.50
0.00
Jan Feb Mar Apr May Jun Jul Aug Sep Oct Nov Dec

FIGURE 5: YEARLY SEASONAL FACTOR OF WAPDA BILLS

By getting Means of monthly Seasonal-Irregulars, Irregular component will be averaged out and we will be left
with uncontaminated Seasonal Indices. Figure 5 is representing yearly cycle of high and low amounts of
WAPDA Bills.

 Deseasonalized Bill amount can be calculated by dividing Wapda Bills by Seasonal Indices.
 Deseasonalized Bill is now free of Seasonal component. Hence, will be used as Dependent Variable in
linear regression.

6|Page
2023-04-0010

Regression Model
Deseasonalized Bill is now free of Seasonal component. Hence, will be used as Dependent Variable in linear
regression. Independent Variables that we are going to use in our Model will be (i)time (ii) new meter (iii)
Dummy Variables of 1st installed Heat Pump and then 2nd Installed Heat Pump.

Model Summary
Adjusted R Std. Error of
Model R R Square Square the Estimate
1 .763a .582 .567 3593.95883
a. Predictors: (Constant), VAR00002, VAR00003,
newmeter, snum

FIGURE 6: REGRESSION MODEL SUMMARY

For further details please refer Exhibit 3.

Regression Equation
y = 5460.193 + 159.97 t + 7463.25 New Meter - 13017.56 1st New Pump – 19444.453 Both New Pumps

 Forecasted Deseasonalized Bill amount corresponding Seasonal Index would be our final Forecasted
WAPDA Bill amount.
Interpretation
R Square: 58% of variability of Dependent Variable is explained by selected regression model.
Time: With 95% certainty, every unit increase in time will increase 160 Rs WAPDA Bill, if other variable is
kept constant
Similarly, only New Meter has contributed in 7463 Rs increase in Monthly Bill and after first Heat Pump and
both Heat Pumps amounts decrease in Monthly Bill are 13017 Rs and 19444 Rs. Respectively.
Forecast Through Classical Decomposition
Months Forecasted Monthly Bill Through Classical Decomposition
Jan 21,571.74
Feb 28,621.95
Mar 19,928.53

7|Page
2023-04-0010

Apr 16,450.89
May 12,700.03
Jun 6,347.36
Jul 8,129.17
Aug 8,213.44
Sep 8,313.59
Oct 7,901.58
Nov 7,449.94
Dec 16,373.65
FIGURE 7: 2001 MONTHLY BILL FORECAST THROUGH CLASSICAL DECOMPOSITON

Refer to Exhibits 4 for further details.

SECTION 5
Compare the forecasts that you have made using regression and classical

decomposition. Explain any difference. Which one would you prefer to use? Why?

Models Comparison
Both models are Forecasting Monthly Bills in entirely different way.

50,000.00
45,000.00
40,000.00
35,000.00
30,000.00 Monthly Bill
25,000.00 Forecasted Monthly Bill Through
20,000.00 Classical Decomposition
15,000.00 Forecasted Bill Through Regression
10,000.00
5,000.00
0.00
1 10 19 28 37 46 55 64 73 82 91 100 109 118

FIGURE 8: COMPARISON OF FORECASTED BILLS WITH ACTUAL BILLS THROUGH CLASSICAL

DECOMPOSITION AND REGRESSION

CLASSICAL DECOMPOSITION REGRESSION FORECAT

8|Page
2023-04-0010

FORECA
From fig 8, shifting with different levels of Not able to shift with different levels of actual
actual bill values quite closely. bill amounts.
MSE is 15,794,021.76 MSE is almost 3 times more 42,834,827.81

58% of variability of Dependent Variable is 39% variability of Dependent Variable is

explained by selected regression model. explained by selected model.
Time variable is coming significant due to Time variable is not coming significant due to
ability of model to shift different levels inability of model to shift different levels

Section 6
Logistic Regression
We need to know probability of having less than 5 family members in the house and using 2000KwH and 4000
KwH. For this a new variable is defined called FamNum, where its value is 0 when there are 5 family members
and 1 when there are less than 5 family members.

Regression Model
We will forecast probability and Odds Ratios with the help of Logistic Regression Model.

Dependent Variable: will be FamNum

Independent Variable: will be electricity usage

Model Summary
-2 Log Cox & Snell Nagelkerke R
Step likelihood R Square Square
1 118.130a .146 .215
a. Estimation terminated at iteration number 6
because parameter estimates changed by less
than .001.

9|Page
2023-04-0010

Variables in the Equation

B S.E. Wald df Sig. Exp(B)
Step 1a elec -.001 .000 11.509 1 .001 .999
Constant .365 .409 .798 1 .372 1.441
a. Variable(s) entered on step 1: elec.

FIGURE 9 & 10: SUMMARY AND OUTPUTS OF BINARY LGISTIC REGRESSION

 In this model our Nagelkerke R Square is .215.

 Beta of Electricity usage is significant and having value of -0.001

Refer to Exhibit 5 for further details.

Regression Equation
eg
Logistic Equation; y=
e g +1

Where, g = b0 + b1 x1

Interpretation
Probabilities:

By inserting values of betas and x1, we get below probabilities.

Electricity Usage 2000 KwH 4000KwH

Probabilities 16.2 % 2.6 %
So, Probabilities of using 2000 KwH and 4000KwH for less than 5 family members are 16.2% and 2.6%
respectively.

Odds Rations:

Odds Ratio = ebi

Odds Ratio = .999

So here, with one unit increase in KwH, Odds will change 0.99 times.

10 | P a g e
2023-04-0010

11 | P a g e