www.ijbcnet.

com

International Journal of Business and Commerce

(ISSN: 2225-2436)

Vol. 1, No.12: Aug 2012[33-46]

Forecasting Unemployment Rates in Nigeria Using Univariate Time Series Models

Louis Sevitenyi Nkwatoh
Yobe State University, Nigeria
Email: Sevinkwatoh@Gmail.Com,
Tel: +2348062218765
Abstract
One of the major problems faced by policy makers is coping with the persistent increase
in the level of unemployment in Nigeria. Thus projecting future unemployment rates is
imperative to policy makers. The main objective of this paper is to search the best
forecasting model among: the trend regression, autoregressive moving average (ARMA),
autoregressive conditional heteroscedasticity (ARCH) and the generalized autoregressive
conditional heteroscedasticity (GARCH) model that could give the best prediction of
future unemployment rates in Nigeria. Applying quarterly data from 1976Q1 to 2011Q4
on unemployment rates, this study evaluated the forecasting performance of the four
competing models using the forecast accuracy criteria such as the root mean squared
error (RMSE), mean absolute percentage error (MAPE), mean absolute error (MAE) and
the Theil's inequality coefficient (U-STATISTICS). The results established that a positive
and significant linear trend factor influenced the time series data. Furthermore, the
Autocorrelation function (ACF) the Augmented Dickey-Fuller (ADF) and Phillips-Perron
tests showed that the time series data was non stationary but was made stationary by
differencing once. Among these models, the empirical study reveals that the mixed
ARIMA/ARCH model could reasonably be used to forecast unemployment rates in
Nigeria specifically in the short-run. Thus for policy implication an ARIMA
(1,1,2)/ARCH (1) is relevant for decision making.
Keywords: JEL: E24, C22, C53
I. Introduction
Unemployment is one of the most challenging problems facing the governments of developing countries.
In Nigeria, statistics show that, the level of unemployment is very high and most prevalent among the
youths, who find it difficult to fulfill their aspirations, assume their economic independence thus failing to
contribute gainfully to the economic development of the Country. The resultant effect of unemployment
in Nigeria can be categorized into two. One, it increases the level of poverty in the country and two, it
causes idleness among the youths, which makes them to engage into undesirable activities, thus wasting
their talents, valuable skills energy and time.
A number of factors among others that have exacerbated the rise of unemployment in Nigeria are: One,
structural factors which have to do with the mismatch between the nature of the educational system and
the needs of the labor market, technical change and the use of capital intensive techniques of production,
the skill mix of the labour force and available job opportunities. Two, Cyclical factors such as the
insufficiency of aggregate local and foreign demand for goods and services. Three, institutional factors
Published by Asian Society of Business and Commerce Research

33

www.ijbcnet.com

International Journal of Business and Commerce

(ISSN: 2225-2436)

Vol. 1, No.12: Aug 2012[33-46]

such as the move by the World Bank and IMF in the 1980s, ordering developing Countries to downsize
their public sector and civil services (ILO Publication, 2005) and also, activities of labor unions and labor
market regulations.
However, various administrative initiatives have been implemented to promote self-dependency and selfreliance that will gainfully generate self-employment, thus reducing the rate of unemployment. For
instance, the introduction of vocational courses in the educational curriculum in 1997, the creation of the
National Directorate of Employment in 1986 charged with the responsibility of promoting skills
acquisition, the National Economic Empowerment and Development Strategies designed in 2004 with
one of its doctrinal values geared towards fighting unemployment, coupled with election promises.
Despite the above initiatives undertaken by the various regimes, reducing unemployment to a desirable
minimum still remains elusive. For instance, Adebayo and Ogunrinola, (2006) and NBS (2010), cited in
Ayoyinka and Oluranti (2011) confirmed that, the rate of open unemployment was 12% in March 2006; it
rose to 19.7% in March 2009 while the rate of underemployment hovered around 19% in 1998 Among the
youths in the 15-24 age cohort, the rate of unemployment is over 40% according to the 2010 edition of the
Labour Force Sample Survey of the National Bureau of Statistics. Thus, signals of future unemployment
rates are necessary for policy makers to plan and strategize before time in order to circumvent the
persistent rise in unemployment levels in the Country. Consequently the main objective of this work is to
identify the best forecasting model among others that can be used to model and forecast future
unemployment rates in Nigeria.
2. Literature Review
Literature on macroeconomic modeling, and forecasting,with the use of historical data from time series
(univariate or multivariate time series) is vast. Modeling unemployment rates like any other
macroeconomic variable has been analyzed traditionally by building econometric models, often related to
stationary time series, ranging from trend analysis, and exponential smoothening to the simple OLS
technique including Autoregressive Integrated Moving Average(ARIMA) models and to the
GeneralisedAutoregressive Conditional Heteroscedastic (GARCH) models (see, Elham et al, 2010; Assis
et al, 2010).
The Box and Jenkins methodology has been extensively used, to project future macroeconomic variables
likewise unemployment rates. Ion and Andriana (2008) used an ARIMA (2,1,2) to forecast future
unemployment rates in Romania. Power and Gasser (2012) established that an ARIMA (1,1,0) model
gave a better forecast for unemployment rates in Canada, while Etuk, Uchendu and Uyodhu (2012)
arrived at an ARIMA (1,2,1) model for Nigeria. The suitability of the ARIMA models for projecting
macroeconomic variables can also be found in studies: Purna (2012), forecasting cement production
output in India; Mordi et al (2006) that analyzed inflation rates in Nigeria; Fatimah and Roslan (1998)
that forecasted cocoa prices in Malaysia etc. Assis et al, (2010) noted that, these models have the
advantage of relatively low research costs when compared with other econometric models, as well as
efficiency in short termforecasting.
Recently, studies have analyzed time series model with the incorporation the Autoregressive Conditional
Heteroscedastic (ARCH) model introduced by Engle (1982). These models have been extended to the
Generalized Autoregressive Conditional Heteroscedastic (GARCH) models leading to more parsimonious
results than ARCH models. This is similar to situations where ARMA models are preferred AR models
Published by Asian Society of Business and Commerce Research

34

www.ijbcnet.com

International Journal of Business and Commerce

(ISSN: 2225-2436)

Vol. 1, No.12: Aug 2012[33-46]

(Assis et al, (2010). Comparing the out-of-sample forecasting accuracy for the United Kingdom
unemployment rate, Floros (2005) established that, though an MA(4) model performed well, while both
MA(1) and AR(4) proved to be the best forecasting models, the MA(4)-ARCH (1) model provided
superior forecasts of unemployment rate in UK. Zhou et al. (2006) proposed a new telecommunication
network prediction model based on non-linear time series ARIMA/GARCH. Their findings suggested that
the ARIMA/GARCH model outperformed the Fractional Autoregressive Integrated Moving Average
(FARIMA) model initially used, in terms of prediction accuracy. Assis et al, (2010) have also
demonstrated that the mixed ARIMA/GARCH model outperformed the exponential smoothing,ARIMA,
and GARCH models when used for forecasting future prices of cocoa beans in Malaysia. A similar
conclusion has been reached by Kamil and Noor (2006) in their approach to forecast the price of
Malaysian raw palm oil using the Autoregressive Conditional Heteroskedasticity (ARCH) model.
However, neither, the ARIMA models nor the ARCH/GARCH models have proven their suitability in
some economies or when used to model some macroeconomic variables. Jaafar,(2006) established that
the Holts method with two parameters was suitable to forecast five major labour force indicators i.e.
labour force, employed, unemployed, unemployment rate and underemployed in Malaysia. Nasir, Hwa
and Huzaifah (2008) used different univariate modelling techniques: the naive with a trend model,
average change model, double exponential smoothing and Holts model, to forecast future unemployment
rates in Malaysia. They used the smallest value of mean square error (MSE) to identify the most suitable
model and evidently concluded that the Holts model outperformed other techniques. However, the
relative advantage of the Holts Method Model over the others is that, recent observations are given
relatively more weighting in forecasting than the older observations.
Gil-Alana (2001) showed that a Bloomfield exponential spectral model gave a feasible result, in lieu of
ARMA models, when modelling UKs unemployment rate. Golan and Perloff (2002) concluded in their
study that, the nonparametric method of forecasting unemployment rates in the U.S outperformed many
other well-known models, even when these models use more information. They however attributed this
result to the nonlinearity in the data generating process.
3. Methodology/Forecasting Models
This work employed unemployment rates data from 1976 to 2011 obtainable from quarterly abstracts and
annual reports of the Central Bank of Nigeria. In this context unemployment rate is regarded as the
percentage of the workforce that is jobless.
Forecasting implies what will happen to the desired event in future, which involves lots of uncertainties. It
involves the process of studying and analysing the past and the present to have a saying over the future
Purna (2012). The mode of analysis can either be judgmental which makes use of personal opinions
(subjectivity) or building structural models which is a quite tasky. This study employed different
econometric time series models like: Trend regression, ARMA, and the mixed ARIMA/GARCH models.
Trend Regression
Regression analysis was used to check if a trend factor exists in the time series data and if significantly
induces changes in unemployment rates.
The regression equation below was used to test for the existence of linear trend factors:
Published by Asian Society of Business and Commerce Research

35

www.ijbcnet.com

International Journal of Business and Commerce

(ISSN: 2225-2436)

Vol. 1, No.12: Aug 2012[33-46]

Where, Yt is the time series data, Trend represents the linear trend factor, and are the coefficients t is the
error term with an assumption that it follows a of white noise (WN) process. The hypothesis of the above
model is such that:
H0: = 0 (No Trend factors exist)
H1: 0 (Trend factors exist)
ARIMA Models
The analysis of ARIMA models follows the Box-Jenkins methodology that combines both the moving
average (MA) and the autoregressive (AR) models. Initially, these models were analyzed by Yule-Walke.
However, a systematic approach that synchronizes both approaches for identifying estimating and
forecasting the models was advanced by Box and Jenkins (1970). The Box-Jenkins methodology begins
with an ARMA(p,q) model which combines both the AR and MA models as follows:
Where, xt represents the explanatory variables, et is the disturbance term. In equation (21), (yt-i) are AR
terms of order p, are MA terms of order q and is a white-noise innovation term. In case of a nonstationary data, the series is differenced (integrated) such: , (d is the number of times a series is
differenced to become stationary; I=d) then the ARMA (p,q) model becomes ARIMA(p,d,q) models
(Auto- regressive Integrated Moving Average of order p,q).
Box-Jenkins ARIMA modeling requires four steps: identification, estimation, diagnostic checking and
forecasting.
1) The identification process starts by testing for the stationary properties of the series. This is done by
analyzing the correlogram of the time series or carrying out a unit root test (Augmented Dickey Fuller
Test and Phillips Perron test).After testing the stationary properties, it is essential to find out highest order
of the ARIMA process. An autoregressive process AR (p) model has partial autocorrelations (PACF) that
truncates at lag p while its autocorrelation functions (ACF) dies off smoothly at a geometric rate. A
moving average process MA (q) has autocorrelation (ACF) that truncates at lag q, while its partial
autocorrelations (PACF) declines geometrically.
Alternative model selection criteria such as Akaike Information Criteria, Schwarz Bayesian Criteria,
Adjusted R2, and Final Prediction Error can be used to verify the order (p,q).
2) After determining the order of p and q the specified regression model is estimated which entails a
nonlinear iterative process of the parameters and. An optimization criterion like least error of sum of
squares, maximum likelihood or maximum entropy is used. An initial estimate is usually used. Each
iteration is expected to be an improvement of the last one until the estimate converges to an optimal one
(Etuk et al, 2012).
3) The fitted model is tested for goodness-of-fit. It can be tested using the above mentioned model
selection criteria. Alternatively, the ACF and PACF obtained from the residual of the specified ARIMA
model as well as the 2 and Ljung-Box Q statistics are diagnostic checking tools. If the residual is free
from all classical assumption of the regression model and stationary then the model is correct (Purna,
2012).
4) The estimated ARIMA model is used to recursively forecast periods ahead.
Consider the general ARMA model:

Published by Asian Society of Business and Commerce Research

36

www.ijbcnet.com

International Journal of Business and Commerce

(ISSN: 2225-2436)

Vol. 1, No.12: Aug 2012[33-46]

Then the forecasted ARIMA model:

Where a, is the intercept term are the parameters of the autoregressive process, are the parameters of
moving average process.
ARCH/GARCH Model
The Autoregressive Conditional Heteroscedastic (ARCH) model was formulated by (Engle, 1982) and
extended to the Generalised Autoregressive Conditional Heteroscedastic (GARCH) model by Bollerslev
(1986). This approach requires a joint estimation of the mean and variance equations. The current
conditional variance a time series depends on the past squared residuals of the process and on the past
conditional variances.
A univariate regression with GARCH (p,q) effects in a polynomial form with a lag operator is represent
as:
Mean Equation:
N (0,) and
Variance Equation:
Where, p is the order of GARCH term and q is the order of ARCH term.
Whereas yt is the endogenous variable and xt exogenous,
is all collected messages up to t-1 period,
and is conditional variance which depends linearly on past squarederror terms and past variances. , are
parameters to be estimated. 1.
ARIMA/GARCH
A combination of the ARIMA (p,d,q) and the GARCH(p,q) are written as be:
4. Analysis of Results
This section began with the analysis of the effect of time on unemployment rate in Nigeria. The
regression result reported below shows that the trend factor has a positive and significant influence on
unemployment rate. That is, time induces a 64.9 percent increase in unemployment rate. This suggests
that the total number of unemployed is more than double as years go by. This result is expected since the
total number of graduates from tertiary institutions is on the increase every year and thus outweighs the
absorptive capacity of the nation. The coefficient of determination (R2) is quite high which indicates that,
the variations in unemployment rate can adequately be explained by a linear trend. The Durbin-Watson
value of 2.25 indicates the absence of serial correlation. The first order auto-regression AR(1) was used to
correction for serial correlation. This is also called the Cochrane-Orcutt correction for serial correlation.
Thus, the trend regression analysis can be used to forecast unemployment rates in Nigeria.
Regression Results Showing Relationship Between Unemployment and Trend
UNEMPLOYMENT = -3.68268 + 0.6498*TREND + 0.79345*AR(1)
Std. Error

6.972

0.2927

6.354

t-Test

-0.528

2.220

6.354

R2 = 0.783; DW = 2.25; F-Statistics = 57.77; p-Value = 0.000

Published by Asian Society of Business and Commerce Research

37

www.ijbcnet.com

International Journal of Business and Commerce

(ISSN: 2225-2436)

Vol. 1, No.12: Aug 2012[33-46]

ARIMA Model
The Box and Jenkins methodology was employed in view of obtaining a forecasting model for
unemployment rate in Nigeria.The initial step in developing a Box-Jenkins model is to determine if the
series is stationary or not. Three important approaches for investigating the stationary properties of a time
series were employed: graphical analysis, the correlogram which analysis the simple autocorrelation
function (ACF) and partial autocorrelation function (PACF) and the unit root test which makes use of the
Augmented
Dickey-Fuller
test
(ADF)
and
Phillips-Perron
test
(PP).

Figure 1: Time Series Plot of Unemployment Rates Series in Nigeria.

The graphical analysis in figure 1 above shows trendy movements; most especially upward trends from
2000 up to 2011. This suggests that the mean of the unemployment rate changes with time. Therefore, the
unemployment rate series is non-stationary.
Figure 2 below, shows the correlogram of the unemployment rate series with a pattern of up to the 36 th
lag. The autocorrelation (ACF) starts with a high value and declines slowly, indicating that the series is
non-stationary. Also the Q-statistic at lag 36has a probability value of 0.000 which is smaller than 0.05
thus not rejecting the null hypothesis that the unemployment rate series is non-stationary. Thus the series
must be differenced for stationarity to occur.
Autocorrelation

Partial Correlation

AC

PAC

Q-Stat

Prob

.|*******

.|*******

0.961

0.961

128.28

0.000

.|*******

.|.

0.922

-0.005

247.42

0.000

.|******|

.|.

0.881

-0.054

357.06

0.000

.|******|

.|.

0.840

-0.032

457.33

0.000

.|******|

.|.

0.797

-0.033

548.41

0.000

.|***** |

.|.

0.757

0.000

631.07

0.000

Published by Asian Society of Business and Commerce Research

38

www.ijbcnet.com

International Journal of Business and Commerce

(ISSN: 2225-2436)

Vol. 1, No.12: Aug 2012[33-46]

.|***** |

.|.

0.717

-0.009

705.88

0.000

.|***** |

.|.

0.678

-0.014

773.32

0.000

.|***** |

.|.

0.640

-0.022

833.77

0.000

.|**** |

.|.

10

0.602

-0.013

887.71

0.000

.|**** |

.|.

11

0.564

-0.020

935.52

0.000

.|**** |

.|.

12

0.528

-0.015

977.64

0.000

.|**** |

.|** |

13

0.509

0.217

1017.2

0.000

.|*** |

*|.

14

0.478

-0.179

1052.4

0.000

.|*** |

.|.

15

0.451

0.014

1084.0

0.000

.|*** |

.|.

16

0.426

0.015

1112.4

0.000

.|*** |

.|.

17

0.406

0.049

1138.5

0.000

.|*** |

.|.

18

0.386

-0.022

1162.1

0.000

.|*** |

.|.

19

0.364

-0.036

1183.4

0.000

.|** |

.|.

20

0.342

-0.019

1202.3

0.000

.|** |

.|.

21

0.317

-0.062

1218.8

0.000

.|** |

.|.

22

0.293

0.002

1232.9

0.000

.|** |

.|.

23

0.268

-0.029

1244.9

0.000

.|** |

.|.

24

0.244

-0.010

1254.8

0.000

.|** |

.|.

25

0.219

0.036

1263.0

0.000

.|*

*|.

26

0.195

-0.107

1269.4

0.000

.|*

.|.

27

0.170

0.029

1274.4

0.000

.|*

.|.

28

0.146

-0.010

1278.1

0.000

.|*

*|.

29

0.111

-0.165

1280.3

0.000

.|*

.|.

30

0.077

-0.016

1281.3

0.000

.|.

.|.

31

0.046

0.010

1281.7

0.000

.|.

.|.

32

0.014

-0.029

1281.7

0.000

.|.

.|.

33

-0.016

-0.017

1281.8

0.000

.|.

.|.

34

-0.048

-0.048

1282.2

0.000

*|.

*|.

35

-0.081

-0.077

1283.4

0.000

*|.

*|.

36

-0.118

-0.070

1286.1

0.000

Figure 2: Correlogram of Unemployment Rate Series

The results of Augmented DickeyFuller (ADF) test and Phillips-Perrons (PP) test on unemployment rate
series is presented on table 2 below. The t-statistic values for both ADF and PP tests are greater than their
corresponding critical values; which reveals the fact that the null hypothesis of the presence of unit root in
the series is not rejected. Hence the series is non stationary.

Published by Asian Society of Business and Commerce Research

39

www.ijbcnet.com

International Journal of Business and Commerce

(ISSN: 2225-2436)

Vol. 1, No.12: Aug 2012[33-46]

Table 1: ADF and Phillip-Perron Test on Unemployment Series.

Augmented Dickey-Fuller test statistic
t-Statistic

Prob.*

Augmented Dickey-Fuller test statistic

0.021503

0.9962

Test critical values:

1% level

-4.027463

5% level

-3.443450

10% level

-3.146455

Phillp-Peron test statistic

Phillips-Perron test statistic

Test critical values:

t-Statistics

Prob.*

0.022666

0.9962

1% level

-4.027463

5% level

-3.443450

10% level

-3.146455

*MacKinnon critical values for rejection of hypothesis of a unit root.

The results of the Augmented Dickey Fuller (ADF) test and Phillips Perron (PP) test presented in table 2
shows that the unemployment series is stationary since the t-statistic values for both ADF and PP tests are
less than their corresponding critical values; which reveals the fact that the null hypothesis of the presence
of unit root in the series is rejected.
Table 2: ADF and Phillip-Perrons Test on Unemployment Series.
Augmented Dickey-Fuller test statistic
t-Statistic

Prob.*

Augmented Dickey-Fuller test statistic

-12.53010

0.0000

Test critical values:

1% level

-4.027959

5% level

-3.443704

10% level

-3.146604

Phillps-Peron test statistic

Adj. t-Stat

Prob.*

Phillips-Perron test statistic

-12.49060

0.0000

Test critical values:

1% level

-4.027959

5% level

-3.443704

10% level

-3.146604

*MacKinnon critical values for rejection of hypothesis of a unit root.

The correlogram in figure 3 shows that the terms (spikes) of the ACF and PACF lie within the band
confidence interval significantly equal to zero at 5 percent. Furthermore, the Q-statistic has a probability
value of 0.953 greater than 0.05 critical value. Therefore the null hypothesis is accepted based on the
coefficient of the correlogram, that the unemployment series is stationary at first difference.
Published by Asian Society of Business and Commerce Research

40

www.ijbcnet.com

Autocorrelation
.|.

.|*

International Journal of Business and Commerce

(ISSN: 2225-2436)
Partial Correlation
.|.

Q-Stat

Vol. 1, No.12: Aug 2012[33-46]

AC

PAC

Prob

-0.042

-0.042

0.2394

0.625

.|*

0.165

0.164

4.0372

0.133

0.062

0.077

4.5745

0.206

.|.

.|*

.|.

.|.

0.068

0.048

5.2196

0.266

.|.

.|.

-0.004

-0.022

5.2214

0.389

.|.

.|.

0.041

0.017

5.4675

0.485

.|.

.|.

0.054

0.055

5.8918

0.552

0.089

0.087

7.0491

0.531

.|*

.|*

.|.

.|.

-0.015

-0.028

7.0832

0.628

.|.

.|.

10

0.021

-0.021

7.1507

0.711

.|.

.|.

11

-0.004

-0.015

7.1531

0.787

12

-0.071

-0.080

7.9013

0.793

*|.

*|.

.|.

.|.

13

0.022

0.020

7.9721

0.845

.|.

.|.

14

0.019

0.040

8.0271

0.888

.|*

.|*

15

0.102

0.108

9.6389

0.842

*|.

*|.

16

-0.087

-0.092

10.815

0.821

.|.

.|.

17

0.054

0.006

11.265

0.842

.|.

.|.

18

0.042

0.064

11.540

0.870

.|.

.|.

19

0.005

0.016

11.543

0.904

20

0.140

0.155

14.715

0.792

.|*

.|*

.|.

.|.

21

0.007

-0.011

14.723

0.837

.|.

.|.

22

0.059

-0.005

15.289

0.850

.|.

.|.

23

0.011

-0.021

15.309

0.883

.|.

.|.

24

0.015

-0.001

15.348

0.910

.|.

.|*

25

0.073

0.076

16.253

0.907

.|.

.|.

26

0.060

0.065

16.864

0.913

.|.

.|.

27

0.029

0.013

17.010

0.931

28

0.171

0.105

22.046

0.779

29

-0.019

-0.029

22.110

0.816

.|*

.|*

.|.

.|.

.|.

*|.

30

-0.011

-0.073

22.132

0.849

.|*

31

0.127

0.153

24.994

0.768

.|*

.|.

.|.

32

-0.020

-0.005

25.065

0.803

.|.

.|.

33

0.026

-0.027

25.191

0.833

.|.

.|.

34

0.003

-0.026

25.193

0.863

.|.

.|.

35

0.065

-0.001

25.968

0.866

.|.

.|.

36

-0.011

0.014

25.993

0.891

Figure 3: Correlogram of unemployment rate after first Difference

Published by Asian Society of Business and Commerce Research

41

www.ijbcnet.com

International Journal of Business and Commerce

(ISSN: 2225-2436)

Vol. 1, No.12: Aug 2012[33-46]

After stationarising the data, an appropriate ARMA (p, q) process was identified using the correlogram.
Different estimated models which fulfilled the criteria of p q 5 were considered and compared,
and the same time models whose parameters were not significant at 5% confidence level were dropped
from the model. The estimation technique began by modelling the conditional mean process by an
autoregressive process AR(1) and the moving average MA(1)with a further analysis of the correlogram of
the residuals. Based on the Akaike criterion and the Schwarz criterion, the ARIMA(1,1,2) model gave a
better fit as shown below.
Table 3: Results of ARMA (1,1, 2) model: (d(Unemployment)
Variable

Coefficient

Std. Error

t-Statistic

Prob.

AR(1)

-0.967567

0.034250

-28.25026

0.0000

MA(2)

-0.906691

0.055668

-16.28735

0.0000

R-squared

0.526898

Mean dependent var

0.002256

Adjusted R-squared

0.523287

S.D. dependent var

1.217538

S.E. of regression

0.840642

Akaike info criterion

2.505621

Sum squared resid

92.57493

Schwarz criterion

2.549085

Hannan-Quinn criter.

2.523284

Log likelihood

-164.6238

Durbin-Watson stat

2.177254

The regression result above indicates that the coefficients of the model are significant based on the tstatistic. The coefficient of determination (R2) is 52.6 %. The DW and the F statistics show the model has
a good fit.
Figure4 shows that the auto-correlation functions (ACF) of the residuals which can also be used to test
fitness of the regression model. It is observed that none of the terms (none of spikes) is exterior to the
confidence intervals and the Q-statistic has a critical probability close to 1. Thus, the residuals may be
assimilated to a white noise process, which implies the model gives a better fit. Thus, it is reasonable
enough to use an ARIMA model to forecast unemployment rates in Nigeria.
Table 5 below shows the Lagrange Multiplier (LM) test. The p-value indicates the presence of an ARCH
effect. This test result give us a firsthand information that unemployment rates in Nigeria can be modelled
using ARCH, GARCH and ARIMA/GARCH models based on the significance of the ARCH and
GARCH term.
Autocorrelatio
Partial
n
Correlation

AC

PAC

Q-Stat

*|.

*|.

-0.095

-0.095

1.2224

.|*

.|*

0.095

0.086

2.4478

Prob

.|.

.|.

0.019

0.036

2.4959

0.114

.|.

.|.

-0.008

-0.011

2.5041

0.286

.|.

.|.

-0.050

-0.058

2.8528

0.415

Published by Asian Society of Business and Commerce Research

42

www.ijbcnet.com

International Journal of Business and Commerce

(ISSN: 2225-2436)

.|.

.|.

-0.029

-0.038

2.9720

0.563

.|.

.|.

0.017

0.022

3.0128

0.698

.|.

.|.

0.030

0.044

3.1407

0.791

.|.

.|.

-0.056

-0.054

3.5971

0.825

.|.

.|.

10

-0.039

-0.063

3.8207

0.873

.|.

.|.

11

-0.042

-0.047

4.0751

0.906

12

-0.113

-0.109

5.9704

0.818

*|.

*|.

.|.

.|.

13

-0.016

-0.023

6.0099

0.873

.|.

.|.

14

-0.023

-0.009

6.0867

0.912

.|*

.|*

15

0.077

0.076

6.9945

0.902

*|.

*|.

16

-0.130

-0.127

9.5735

0.793

.|.

.|.

17

0.027

-0.023

9.6881

0.839

.|.

.|.

18

0.010

0.022

9.7044

0.882

.|.

.|.

19

-0.027

-0.015

9.8158

0.911

20

0.117

0.123

11.994

0.848

.|*

.|*

.|.

.|.

21

-0.024

-0.025

12.085

0.882

.|.

.|.

22

0.030

-0.022

12.235

0.908

.|.

.|.

23

-0.019

-0.038

12.292

0.931

.|.

.|.

24

-0.015

-0.018

12.329

0.950

.|.

.|.

25

0.051

0.059

12.759

0.957

.|.

.|.

26

0.037

0.057

12.994

0.966

.|.

.|.

27

0.005

0.007

12.999

0.977

28

0.149

0.107

16.784

0.915

29

-0.052

-0.033

17.259

0.925

.|*

.|*

.|.

.|.

.|.

*|.

30

-0.042

-0.070

17.574

0.936

.|*

31

0.111

0.156

19.737

0.901

.|*

.|.

.|.

32

-0.045

0.004

20.098

0.914

.|.

.|.

33

0.002

-0.016

20.099

0.934

.|.

.|.

34

-0.022

-0.022

20.183

0.948

.|.

.|.

35

0.040

0.001

20.479

0.956

.|.

.|.

36

-0.039

0.013

20.763

0.964

Vol. 1, No.12: Aug 2012[33-46]

Figure 4: The Correlogram of Residuals.

Table 4: ARCH LM Test on ARIMA (1,1, ) Residuals.
F-statistic

20.28684

Prob. F(2,129)

<0.01

Obs*R-squared

43.88374

Prob. Chi-Square(2)

<0.01

Published by Asian Society of Business and Commerce Research

43

www.ijbcnet.com

International Journal of Business and Commerce

(ISSN: 2225-2436)

Vol. 1, No.12: Aug 2012[33-46]

The regression result below shows that the coefficients of the model are significant as well the ARCH
term. Therefore ARCH model can be a suitable model. This study rejected the GARCH model because all
the different GARCH terms used in the different estimation models were insignificant
Table 5: Results of ARMA (1,1, 2)/ARCH(1) model
Variable

Coefficient

Std. Error

z-Statistic

Prob.

AR(1)

-0.942958

0.006804

-138.5906

0.0000

MA(2)

-0.937420

0.009176

-102.1596

0.0000

Variance Equation
C

0.095794

0.009674

9.902176

0.0000

RESID(-1)^2

2.157868

0.332573

6.488409

0.0000

R-squared

0.517637

Mean dependent var

0.002256

Adjusted R-squared

0.513955

S.D. dependent var

1.217538

S.E. of regression

0.848830

Akaike info criterion

1.708744

Sum squared resid

94.38702

Schwarz criterion

1.795671

Hannan-Quinn criter.

1.744068

Log likelihood

-109.6314

Durbin-Watson stat

2.246583

The results of LM test on table 5 indicate that the residuals do not show any ARCH effect which implies
that the ARCH (1) model has adequately captured the persistence in volatility of the series. Hence, the
ARIMA (1,1,2)/ARCH (1) model can reasonably be used to model and forecast unemployment rates in
Nigeria.
Table 6: ARCH LM Test on ARCH(1)residuals.
F-statistic

0.140101

Prob. F(1,130)

0.7088

Obs*R-squared

0.142103

Prob. Chi-Square(1)

0.7062

Model Selection Criteria

Four model selection criteria: the Root mean squared error (RMSE), mean absolute percentage error
(MAPE), mean absolute error (MAE) and Theil's inequality coefficient (U-Statistics) were used to
determine the best forecasting model from the three identified models that is the trend regression,
ARIMA(1,1,2) and ARIMA (1,1,2)/ARCH (1) models. The table below shows that,from the three
models considered, the ARIMA (1,1,2)/ARCH(1) gave the best forecasting model for unemployment
rates in Nigeria.

Published by Asian Society of Business and Commerce Research

44

www.ijbcnet.com

International Journal of Business and Commerce

(ISSN: 2225-2436)

Vol. 1, No.12: Aug 2012[33-46]

Table 7: Model Selection Criteria..

Criteria

Trend

RMSE

14.51615

5.408114

5.321303

MPAE

276.9532

70.27601

70.25185

MAE

12.95547

3.988209

3.955172

U-Statistics

0.439882

ARIMA (1,1,2)

ARIMA (1,1,2)/ARCH (1)

0.314895

0.305862..

5. Conclusion
Projecting future unemployment rates like any other macroeconomic variable should be an important
application to economists as well as policy makers. This paper investigated the different univariate time
series models used for forecasting unemployment rates in Nigeria, namely trend regression analysis,
ARIMA, GARCH, and the mixed ARIMA/GARCH models. Specifically, the forecasting techniques were
compared based on the following criteria: Root Mean Square Error (RMSE), Mean Absolute Error
(MAE) and Mean Absolute Percent Error (MAPE). Though all the models could be used for projection
based on the significance of the parameters and the fitness of the models, the model selection criteria
showed that the ARIMA/ARCH model outperformed the trend regression and the ARIMA models.
Basically the results showed that unemployment rates in Nigeria could be modelled and predicted using
an ARIMA (1,1,2)/ARCH(1) model. This results contradicts the findings of Etuk et al (2012) whose
results showed that an ARIMA (1,2,1) could be used to forecast unemployment rates in Nigeria using
monthly data from 1999 to 2008. This result is however possible since forecasting any macroeconomic
variable may be affected by changes in time horizon (periods) as well as the sample size of the data.
A possible recommendation is that, since short-run projections are better, there should be continuous
investigation of an appropriate model that could be used to project future unemployment rates. Further
research on both simple and complex econometric techniques is imperative so that long-run
unemployment rates could as well be projected.
Reference
Assis, K., Amran, A.andRemali, Y. (2010): Forcasting Cocoa Beans Pricis Using Univariate Time Series
Models. International Refereed Research Journal www.researchersworlld.com Vol. I, Issue
1Pp 71-80.
Bollerslev, T. (1986); Generalized Autoregressive Conditional Heteroscedasticity. Journal of
Econometrics, 31(3): 307-327.
Celia, F., Ashish, G., Amar, R., and Les, S. (2003): Forecasting womens apparel sales using
mathematical modelling. International Journal of Clothing, Science and Technology, 15(2):107125
Elham, K.,Masoumeh, Z. and Ebrahim, B. (2010): Prediction of added value of agriculturalsubsections
using artificial neural networks: Box-Jenkins and Holt-Winters methods. Journalof Development
and Agricultural Economics, 2(4): 115-121. URL:
http://www.academicjournals.org/JDAE/PDF/Pdf2010/Apr/Kahforoushan%20et%20al.pdf

Published by Asian Society of Business and Commerce Research

45

www.ijbcnet.com

International Journal of Business and Commerce

(ISSN: 2225-2436)

Vol. 1, No.12: Aug 2012[33-46]

Engle, F. (1982): Autoregressive Conditional Heteroscedasticity with Estimates of theVariance of United

Kingdom Inflation. Econometrica, 50(4): 987-1008.
Ette H., Uchendu B, and Uyodhu V (2012): Arima Fit to Nigerian Unemployment Data. Journal of
Basic and Applied Scientific Research, 2(6)5964-5970 ISSN 2090-4304.
Fatimah, M. and Roslan, A. (1986): Univariate approach towards cocoa price forecasting.The
Malaysian Journal of Agricultural Economics, 3: 1-11.
Floros, C. (2005): Forcasting the UK Unemployment Rate: Model Comparisons.International Journal
of Applied Econometrics and Quantitative Studies.Vol.2-4. Pp 55-72.
Gil-Alana, A. (2001): A fractionally Integrated Exponential Model for UK Unemployment, Journal of
Forecasting, 20, 329- 340.
Golan, A. and Perloff, J.(2002): Superior Forecasts of the U.S. Unemployment Rate Using a
Nonparametric Method. UDARE Working Papers, Paper 956. http://repositories.cdlib.org/are
ucb/956
Ion, D. and Andriana, A. (2008): Modeling unemployment Rate Using Box-JienkingsProcedure.Journal
of Applied Quantitative Methods.Vol 3. No 2. Pp 156-166.
Jaafar,

J. (2006):PeramalanIndikatorUtamaTenagaBuruh.
Department of Statistics Malaysia

Research andDevelopment

Division,

Kamil, A. and Noor, A. (2006): Time series modeling of malaysian raw palm oil price:
Autoregressive conditional heteroskedasticity (ARCH) model approach.Discov. Math., 28:19-32.
Nasir, M., Hwa, K. and Huzaifah, M (2006): An Initial Study on the Forecast Model for Unemployment
Rate. Workshop Review of the Labour Force/Migration Survey ,Putrajaya.
Power, B. and Gasser, K. (2012): Forecasting Future Unemployment Rates. ECON 452 First Report.
Purna, C. (2012): Use of Univariate Time Series Models for Forecasting Cement Productions in
India.International Research Journal of Finance and Economics.ISSN 1450-2887.Issue 83.Pp
167-179.
Telesca, L., Bernardi, M. and Rovelli, M. (2008): Time-scaling analysis of lightning in Italy.
Commun.Nonlinear Sci. Numer. Simulation, 13: 1384-1396.
Zhou, B., He, D., and Sun, Z. (2006): Modelling and Simulation Tools for Emerging
Telecommunication Networks. Springer US: 101-121. ISBN 978-0-387-32921-5.

Published by Asian Society of Business and Commerce Research

46