European J ournal of Scientific Research

ISSN 1450-216X Vol.58 No.1 (2011), pp.30-37

EuroJ ournals Publishing, Inc. 2011
http://www.eurojournals.com/ejsr.htm

Forecasting Electricity Generation in Nigeria using
Univariate Time Series Models


S. O. Adams
Corresponding Author, Department of Statistics, University of Abuja, Abuja, Nigeria
E-mail: adamsamuel22003@yahoo.com

R. O. Akano
Department of Statistics, University of Abuja, Abuja, Nigeria

O. J . Asemota
Department of Economic Engineering, Kyushu University, Fukuoka, Japan


Abstract

This paper fit a univariate time series model to the average amount of electricity
generated in Nigeria between 1970 and 2009 and provides ten years forecast for the
expected electricity generation in Nigeria. The Box-J enkins Autoregressive Integrated
Moving Average (ARIMA) models are estimated and the best fitting ARIMA model is
used to obtain the post-sample forecasts. The fitted model was ARIMA (3,2,1),with the
Normalized Bayesian Information Criteria (BIC) of 13.906, stationary R
2
=0.69 and
Maximum likelihood estimate of 411.55. The model was further validated by Ljung-Box
test (Q
14
=6.404 and p>.10) with no significant autocorrelation between residuals at
different lag times. Finally, ten years forecast was made, which showed a pick in the
average electricity generation with estimated value as 3088.22 in the year 2011.


Keywords: ARIMA, Autocorrelation, Partial autocorrelation, Serial dependence, Ljung-
Box-Pierce, PHCN.

1. I ntroduction
Electricity generation is one of the seven point agenda of the present Nigeria government. Electric
power industry in many countries all over the world is moving from a centralized operational approach
to a competitive one, Nogales and Contreras (2002). In 2010, the Nigerian government took a bold step
in greater involvement of the private sector in power generation in Nigeria. The Nigerian power sector
operates well below its estimated capacity, with power outages being a frequent occurrence. In 2003,
total installed electricity capacity was 5.9 gigawatts(GW), while total consumption was 14.5 Bkwh.
According to Power Holding Company of Nigeria (PHCN), the electric demand in February 2006 was
7,600 megawatts (MW), but actual generation capability was 3,600 MW. The discrepancy between
electricity demand and actual generation is mostly due to low water levels and inadequate plant
maintenance. During year 2005, electricity generation capacity fluctuated between 2,600 MW and
3,600 MW. The electricity generated by hydropower stations at Kainji, J ebba, and Shiroro has been
bedeviled by insufficient water. The Lagos Egbin, Delta, and Port Harcourt Afam plants are also
operating at below capacity due to poor maintenance.
Forecasting Electricity Generation in Nigeria using Univariate Time Series Models 31

Only 40% of Nigerians have access to electricity, the majority of who are concentrated in urban
areas. Despite endemic blackouts, customers are billed for services not rendered, partially explaining
Nigeria's widespread vandalism, power theft and PHCN's problems with payment inadequacy. Some
industries have collapsed due to the epileptic power supply in the country, Nigerias Bureau of Public
Enterprises (BPE) hopes to see increased stability in Nigerias electricity sector once the ongoing
privatization of PHCN fully takes effective.
Despite the importance of Electricity generation to the Nigerian economy, the literature is yet
still scanty with published articles on Nigeria Electricity generation. This paper seeks to provide an
appropriate ARIMA model to the data on electricity generation in Nigeria from 1970 - 2009. We
choose this methodology because there is little or no research on electricity generation in Nigeria using
the Box J enkins analysis (see section 2). ARIMA model is used because of its generality; it can handle
many series regardless of stationarity or not, with seasonal or without seasonal elements. The principle
of parsimony of ARIMA model is also taken into consideration. The rest of the paper is structured as
follows; section 2 presents the modeling and methodology, empirical results are presented in section 3,
section 4 dealt with forecasting with the fitted model. The last section discussed the results obtained
and suggested some recommendations.


2. Methodology
2.1. Model Specification
The model used in this study is the ARIMA proposed by Box and J enkins (1976). The preliminary test
for stationarity and seasonality of the data was conducted in which differences (d) as well as natural log
were taken. After the stationarity of the series was attained, ACF and PACF of the stationary series are
employed to select the order p and q of the ARIMA model. At this stage, different candidates model
manifested and their parameters are estimated using the maximum likelihood method. Based on the
principle of parsimony and model diagnostic tests, we obtained the best fitting ARIMA model.

2.2. Source of Data
The data used in this research work was extracted from the Central Bank of Nigeria statistical bulletin
(December, 2007) and materials from Power holding Company Nigeria (PHCN). We obtained data on
electricity power generation in Nigeria (1970 2009). The choice of this range is due to availability of
the data from the source.

2.3. Method of Estimation: ARI MA Methodology
The Box-J enkins model building techniques consists of the following four steps:
Step 1: Preliminary Transformation: If the data display characteristics violating the
stationarity assumption, then it may be necessary to make a transformation so as to produce a series
compatible with the assumption of stationarity. After appropriate transformation, if the sample
autocorrelation function appears to be nonstationary, differencing may be carried out.
Step 2: I dentification: If

t
y
is the stationary series obtained in step 1, the problem at the
identification stage is to find the most satisfactory ARMA (p,q) model to represent

t
y
. Box - J enkins
(1976) determined the integer parameters (p,q) that govern the underlying process

t
y
by examining
the autocorrelations function (ACF) and partial autocorrelations (PACF) of the stationary series,

t
y
.
This step is not without some difficulties and involves a lot of subjectivity. It does on occasion happen
that evidence examined at this stage may not point clearly in the direction of a single model (Salau,
1998). Hence, it is useful to entertain more than one structure for further analysis. Salau (1998) stated
that this decision can be justified on the ground that the objective of the identification phase is not to
32 S. O. Adams, R. O. Akano and O. J . Asemota
rigidly select a single correct model but to narrow down the choice of possible models that will then be
subjected to further examination.
Step 3: Estimation of the model: This deals with estimation of the tentative ARIMA model
identified in step 2. The estimation of the model parameters can be done by the conditional least
squares and maximum likelihood.
Step 4: Diagnostic checking: Having chosen a particular ARIMA model, and having estimated
its parameters, the adequacy of the model is checked by analyzing the residuals. If the residuals are
white noise; we accept the model, else we go to step 1 again and start over.


3. Empirical Result
In this section the ARIMA modeling strategy discussed in section 2.3 is applied to analyze the average
electricity generation data. In this framework, model building commences with the examination of the
plot of the series, the second logged difference plot, the sample plot of the autocorrelations (ACF),
Partial autocorrelations (PACF), model description and forecast value using the fitted model. As in the
first step of the Box- J enkins, we tested for stationary in the data on the average electricity generation
in Nigeria (1970 2009), (see fig1). An examination of Fig 1 clearly revealed that non stationarity is
inherent in the data, after second differencing and taking Natural logarithm of the series, (see fig 2), we
observed that the data on the chart was stationary. From a close observation of the ACF and PACF of
the second logged differenced series, we noticed that the ACF cuts off at lag 1 but with significant
spikes at lags 4. For the PACF plot, it is observed that it cuts of at lags 1 and 3 with a significant spike
at lag 2. This implies that the stochastic process that generated the second logged differenced of the
average electricity generated data is an ARMA model which has at most an MA (3) component. Hence,
a number of possible models manifest themselves, these are ARMA (1,||1,2,3), ARMA (3,||1,2,3). i.e.,
ARIMA(1,2,1), ARIMA(1,2,2), ARIMA(1,2,3), ARIMA(3,2,1), ARIMA(3,2,2) and ARIMA(3,2,3).
We proceeded to further statistical analysis with the six possible models. We summarized the results in
Table 1.

Figure 1: Time Series plot of Electricity generated in Nigeria, (1970 2009)







Forecasting Electricity Generation in Nigeria using Univariate Time Series Models 33

Figure 2: Time Series plot of Second Logged difference



Figure 3: ACF of the second logged difference of Electricity generated in Nigeria



Figure 4: PACF for second logged difference of Electricity generation in Nigeria




34 S. O. Adams, R. O. Akano and O. J . Asemota
Table 1: Model Description

ARIMA
STRUCTURE
Parameter
Estimate
P - Value
Stationary
R
2

Likelihood &
BI C
Standard
Error of
estimate
Q- Statistics
1. ARIMA (1,2,1)
AR{1} =-0.194
MA{1} =0.998
0.324
0.852
0.58
371.183
BIC =13.971
0.193
5.310
12.763(0.690)
2. ARIMA(1,2,2)
AR{1} =-0.997
MA{1} =-0.085
MA{2} =0.881
0.000
0.835
0.032
0.50
277.182
BIC =14.267
0.117
0.406
0.393
8.911(0.882)
3. ARIMA (1,2,3)
AR{1} =-1.00
MA{1} =-0.083
MA{2} =0.853
MA{3} =-0.052
0.000
0.835
0.013
0.800
0.51
401.204
BIC =14.267
0.046
0.393
0.325
0.203
9.580(0.792)
4. ARIMA (3,2,1)
AR{1} =-0.478
AR{2} =-0.504
AR{3} =-0.370
MA{1} =0.985
0.017
0.017
0.063
0.094
0.69
411.555
BIC =13.906
0.189
0.200
0.192
0.520
6.404(0.955)
5. ARIMA(3,2,2)
AR{1} =-0.686
AR{2} =-0.580
AR{3} =-0.446
MA{1} =0.753
MA{2} =0.246
0.150
0.024
0.045
0.974
0.967
0.68
275.8
BIC =14.037
0.464
0.244
0.214
23.157
5.929
6.352(0.932)
6. ARIMA(3,2,3)
AR{1} =-0.580
AR{2} =-0.889
AR{3} =-0.472
MA{1} =0.803
MA{2} =-0.315
MA{3} =0.509
0.196
0.016
0.038
0.869
0.777
0.847
0.67
275.302
BIC =14.143
0.438
0.349
0.217
4.829
1.104
2.614
1.595(.998)
Notes: and denote significant at the 1% and 10% levels respectively. Figures in parenthesis also denote P-values.

From Table 1, ARIMA structure 4 seems to be the most competitive model. The parameter
estimates are all significant, the value of its stationary R
2
is the highest and the Q statistics are also
insignificant. The most important summary statistics for measure of goodness of fit are the R
2
,
likelihood function (for maximum likelihood estimation), standard error of estimate and the Q statistic.
For a well fitted model, the Q statistic is expected to be statistically insignificant. Another important
criterion for checking the adequacy of a fitted model is the Normalized Bayesian Information Criteria
(BIC). When considering several ARMA models, we choose the one with the lowest BIC. Based on
these four important statistics and BIC, ARIMA structure 4 i.e. ARIMA (3,2,1) seems to provide the
best satisfactory fit to the second logged differenced electricity generation series. This model has the
highest likelihood function and the smallest standard error of estimate among all the ARIMA structures
considered. Besides, the Q statistics is statistically insignificant suggesting that the residuals do not
suffer from autocorrelation.


4. Forecasting with the Fitted Models
In time series modeling researchers are motivated by the desire to produce a forecast with minimum
error as possible. In this section, we assess the forecasting performance of Box-J enkins models. The
traditional Box-J enkins approach is general and can handle effectively many series encounter in reality.
Besides, previous research has demonstrated that the Box-J enkins forecast out performs the Holt-
Winters and stepwise auto regression forecasts, (Newbold and Granger, 1974). In addition, Naylor,
T.H. et al. (1972) also showed the Box-J enkins method give better forecasts than traditional
econometric methods. Forecast from ARIMA model can be computed directly from the ARIMA model
equation by replacing, (1) future values of the error term by zero (2) future values of the
t
y
by their
Forecasting Electricity Generation in Nigeria using Univariate Time Series Models 35

conditional expectation (3) present and past values of
t
y
and
t
c
by their observed values. For example,
the ARIMA (1,1,1) model ;
1 1
)
2 1
(
1
+ +

t

t

t
y
t
y
t
y
t
y
, may be written as
1 1 2 1 1
) (1

+ + + =
t t - t - t t
y y y
has MMSE forecast at time
N computed as;

> +
= +
= + +
=

3 ) 2 ( ) 1 ( ) 1 (
2 ) 1 ( ) 1 (
1 ) 1 (
) (
1 1
1 1
1 1 1 1
i i | i |
i | |
i c | | |
i
N N
N N
N N N
N
y y
y y
y y
y

Where use is made of the parameter estimates of
1 1

u |
and the observed residual
N
c
, see
(Chatfield, 2000) for more detailed discussions. By applications of the procedures discussed above, we
computed one-step ahead forecasts for the fitted mode, i.e. ARIMA (3,2,1). These forecasts and their
95% confidence interval i.e. Lower confidence limit (LCL) and upper confident limit (ULC) for 10
years (i.e. 2010 2019) are summarized in Table 2, while Figure 5 depicts the observed and forecast
plots of electricity generation in Nigeria.

Table 2: Forecasted values with the Fitted Models

YEARS LCL FORECAST UCL
2010 1451.55 2423.91 3825
2011 1711.87 3088.22 5175
2012 1548.03 2847.73 4843
2013 1230.00 2308.77 3987
2014 1171.22 2387.21 4381
2015 1206.43 2629.10 5061
2016 1144.15 2591.98 5127
2017 1012.28 2373.3 4807
2018 937.91 2314.77 4854
2019 905.74 2358.66 5125

Figure 5: The plot of the observed and forecast value of Electricity generation in Nigeria



36 S. O. Adams, R. O. Akano and O. J . Asemota
The result presented above indicates that the average electricity generation in Nigeria will
decrease in the long run, expect for the year 2011, where a slight increment is noticed.


5. Conclusions and Recommendations
The epileptic nature of electricity generation in Nigeria has become unbearable to most Nigerians,
especially in the big cities. It has posed a constant threat to the growth of the countrys economy.
Furthermore, the present leadership as well as the incoming Government needs to take the issue of
electricity generations seriously.
Forecasting the amount in megawatt the electricity generation can help the Government to take
effective measure to handle any unexpected situation. This study is among the few in Nigeria on this
area.
The paper examined the appropriate model that fits the aggregate electricity generated in
Nigeria between 1970 and 2009. It was discovered that ARIMA (3,2,1) is the most suitable model for
the series with the Normalized Bayesian Information Criteria (BIC) of 13.906, stationary R
2
=0.69 and
Maximum likelihood estimate of 411.55 and the Ljung-Box test (Q
14
=6.404 and p>.10) was also
estimated. The ARIMA model revealed that the average electricity generation in Nigeria will reduce
further, with only a slight increment of about 3088.22 megawatt this year, (i.e. 2011).
Although a little increment is forecasted, in year 2011, the predicted megawatt is still not
adequate for a population of over 150 million.
We therefore recommend that Government should as a matter of urgency hasten the
privatization process so that more electricity can be generated given the available numerous natural
resources in the country.
To date, there is no agreement on the actual figure expended in the energy sector from 1999 to
2007 this in itself is very worrisome, in the meantime, we urge the National Assembly to step up its
oversight functions so as to find answers to the whereabouts of the $16 billion that the last
administration claimed to have invested in the power sector and move the nation forward by ensuring
that recent promises by the current administration remain a marked difference from previously failed
promises.


References
[1] Asemota, O.J . (2010). Modelling Nigerias crude oil Exports: State Space versus ARIMA
model, Manuscript Submitted for Publication.
[2] Ayodele A.S. (1998) Energy Crises in Nigeria: The Case of Electric Energy Market in
Bullion, 22 (4), pp. 19-23
[3] Brockwell, P.J . and Davis, R.A. (1996) (Introduction to Time series and forecasting Springer,
New York. Section 3.3 and 8.3
[4] Central Bank of Nigeria (2008) Statistical Bulletin, 14(2)
[5] Central Bank of Nigeria (1999), Annual Report and Statement of Account. Dec, pp 57 - 58,
129.Nigeria.
[6] Chatfield, C. (2000). Time-Series Forecasting, Chapman & Hall/ CRC.
[7] Dickey, D.A. and Fuller, W.A. (1981) Likelihood Ratio Statistics for Autoregressive Process.
Econometrics, 49(3), pp. 1057-1072
[8] Durbin, J . and Koopman, S.J . (2001). Time Series Analysis by State Space Methods. Oxford
University Press.
[9] Ezenwe, U. (1988) The Limits of Privatization in a Developing Economy Privatization of
Public Enterprises in Nigeria. Ibadan: Nigerian Economic Society Seminar Series.
[10] Fawibe, O. (1977): Liberalizing Nigerian Energy and Mineral Industries. International
EnergyServices Ltd., Lagos. Nigeria.
Forecasting Electricity Generation in Nigeria using Univariate Time Series Models 37

[11] Harvey, A.C.(1993) Time Series Models, 2
nd
Edition, Harvester Wheatsheaf, sections 3.3 and
4.4.
[12] Murray, R. S. Theory and Problems of Statisticsand Schuams Outline Series, 1st ed.,
McGraw Hill, Book Co., U. S. A.
[13] National Electric Power Authority (1999).NEPA, Review, J an - Mar. PRO Division, National
ElectricPower Authority, Nigeria.
[14] National Electric Power Authority (1999).NEPA, Review, Dec., PRO Division National
ElectricPower Authority, Nigeria.
[15] Naylor, T.H., Seaks, T.G. and Wichern, D.W. (1972). Box-J enkins Methods: An alternative to
Econometric Models, International Statistical Review, Vol. 40, No. 2, 123-137.
[16] Newbold, P. and Granger, C.W.J . (1974). Experience with Forecasting Univariate Time Series
and the Combination of Forecasts, J ournal of the Royal Statistical Society, Series A, Vol.137,
131-165.
[17] Olowo, B. (2002): NEPA Meets The Challenge.African Review.April, pp 42. U.K.
[18] Osakwe, E. C. N. (1979): Towards a Comprehensive. Energy Policy for
[19] Nigeria.Proc. Energy Policy Conf., National Policy Development Center.Nigeria.
[20] Osamgbi, I. (2001): Energy. This Day Publications, Vol. 7, No. 2424.pp 27. Nigeria.
[21] Salau, M.O. (1998). Arima Modelling of Nigerias Crude Oil Exports, AMSE, Modelling,
Measurement & Control, Vol. 18, No. 1, 1 20.
[22] Sharma, V. C. and Sharma, A.( 1981). Development of Hydropower in Nigeria.Energy.Vol.
6.pp 475 478. U.K.




Copyright of European Journal of Scientific Research is the property of EuroJournals, Inc. and its content may
not be copied or emailed to multiple sites or posted to a listserv without the copyright holder's express written
permission. However, users may print, download, or email articles for individual use.