Modeling Tea Prices and Their Volatility in Kenya

Download as pdf or txt
Download as pdf or txt
You are on page 1of 45

FORECASTING TEA PRICES AND THEIR

VOLATILITY IN KENYA

Robin Ochieng Otieno


SCT-325-002/2019

Department of Mathematics
Multimedia University of Kenya
MSC(Statistics)

October 2022

Supervisors:
Dr. Wycliffe Cheruiyot
Dr. Antony Karanja
October 2022 Supervisors: Dr. Wycliffe C
Robin Ochieng Otieno SCT-325-002/2019 (Multimedia University of Kenya) 1 / 45
Introduction: Background

Tea in Kenya
• Tea has become an essential contributor to the economies of
producing countries such as Kenya, China, India, and Sri Lanka. Tea
originated in China, and it has now advanced to a multi-billion-dollar
global industry.
• In 2019, the tea sector accounted for 25 percent of the foreign
exchange earnings and 1.5 percent of the Gross Domestic Product in
Kenya KIPPRA (2020).

October 2022 Supervisors: Dr. Wycliffe C


Robin Ochieng Otieno SCT-325-002/2019 (Multimedia University of Kenya) 2 / 45
Introduction: Background

Tea in kenya
Globally, Kenya falls in third place after China and India in tea
production with a market share of 9% KIPPRA (2020).
Kenya is the largest exporter of tea in the world accounting to 20
percent of the world tea exports Monroy et al. (2019).
The Kenya tea sector contributes 26 percent of Kenya’s export
earnings as Kenya exports tea to Pakistan, Egypt, the United
Kingdom, and others Muthamia and Muturi (2015).

October 2022 Supervisors: Dr. Wycliffe C


Robin Ochieng Otieno SCT-325-002/2019 (Multimedia University of Kenya) 3 / 45
Introduction: Basic Concepts

Volatility Definition and Measurement


We define volatility as the irregular rising and falling observed in some
events over time.
In forecasting literature it is referred to as the conditional variance of
the underlying return.
The sample standard deviation is used as its measure;
r Pn
2
i=1 (xi − µ)
σ= (1)
N

October 2022 Supervisors: Dr. Wycliffe C


Robin Ochieng Otieno SCT-325-002/2019 (Multimedia University of Kenya) 4 / 45
Introduction: Basic Concepts continued..

Volatility Clustering
Volatility clustering behavior refers to the tendency for large changes
in price to be followed by large changes and small changes to be
followed by small changes Rachev et al. (2011).
It is determined by enumerating the Autocorrelation function.
Suppose Xt is a stationary time series, the ACF for the series is
defined as:

Corr (Xt , Xt−k ) γ(k)


ρk = p = (2)
Var (Xt )Var (Xt−k ) γ(0)

October 2022 Supervisors: Dr. Wycliffe C


Robin Ochieng Otieno SCT-325-002/2019 (Multimedia University of Kenya) 5 / 45
Introduction: Basic Concepts continued..

Stationarity in Time Series


Stationarity is a critical assumption in time series models.
Stationarity implies homogeneity in the sense that the series behaves
in a similar way regardless of time.
This means that its statistical properties ie the mean, variance and
the autocovariance of the series do not change over time that is
independent of time.

October 2022 Supervisors: Dr. Wycliffe C


Robin Ochieng Otieno SCT-325-002/2019 (Multimedia University of Kenya) 6 / 45
Introduction: Basic Concepts continued..

Differencing in Time Series


The most common way to make non-stationary time series data
stationary is by differencing.
Differencing is very useful in eliminating the trend and seasonal
effects. For seasonal data, first order differencing is generally
adequate to achieve stationarity.
Let Xt = X1 , X2 , ...Xn be a non staionary time series. Then stationary
time series is obtained by:

∇Xt = Xt − Xt−1 (3)


This is called a first order difference.

October 2022 Supervisors: Dr. Wycliffe C


Robin Ochieng Otieno SCT-325-002/2019 (Multimedia University of Kenya) 7 / 45
Introduction: Basic Concepts continued..

Returns
Directly analyzing financial prices is difficult since they are correlated
and their variances typically rise with time; hence, we need returns
for analysis.
Let Yt and Yt−1 be the current and previous monthly prices of tea
then the returns can be defined as

Yt
rt = log (4)
Yt−1

October 2022 Supervisors: Dr. Wycliffe C


Robin Ochieng Otieno SCT-325-002/2019 (Multimedia University of Kenya) 8 / 45
Literature Review

Literature Review
• According to Gesimba et al. (2005), the tea industry is facing a
number of issues. The most challenging problem, he argues, is the
unpredictability of tea export prices, which may be overcome by
regulating demand and supply in the global market.
• The ultimate purpose of volatility analysis is to explain why there is
volatility. The GARCH models are intended to study volatility in order
to make financial choices such as risk evaluations.

October 2022 Supervisors: Dr. Wycliffe C


Robin Ochieng Otieno SCT-325-002/2019 (Multimedia University of Kenya) 9 / 45
Literature Review continued..

Literature Review
• Chinomona (2009) modeled South African inflation data using the
ARIMA and GARCH models; she picked the
ARIMA(1, 1, 0) ∗ (0, 1, 1)12 model from the ARIMA models and
GARCH (1, 1) from the ARCH-GARCH models.
• Because of its capacity to capture fluctuations in the series, GARCH
(1, 1) was chosen as the best model over SARIMA(1, 1, 0) ∗ (0, 1, 1)12 .
• Another research by Fwaga et al. (2017) employed both the GARCH
and the EGARCH models to model Kenyan inflation. The model with
the lowest AIC and BIC values was chosen as the best model.

October 2022 Supervisors: Dr. Wycliffe C


Robin Ochieng Otieno SCT-325-002/2019 (Multimedia University of Kenya) 10 / 45
Statement of the Problem

Problem Statement
• The ultimate goal of price volatility analysis is to explain why prices
fluctuate. Price volatility can be caused by shocks to demand and
supply, which can be caused by changes in climatic conditions, such
as extreme cold or hot weather.
• Because they take volatility clustering into account, GARCH models
are appropriate for forecasting volatility. Their ability to reliably
estimate time-varying conditional variances also enables them to
forecast return variances and covariances.

October 2022 Supervisors: Dr. Wycliffe C


Robin Ochieng Otieno SCT-325-002/2019 (Multimedia University of Kenya) 11 / 45
Objectives of the Study

The Main Objective

The main aim of this study was to forecast tea prices and their
volatility in Kenya.
The Specific Objectives
1 To determine the optimal SARIMA and GARCH models on the monthly
price and returns of tea in Kenya.
2 To diagnose the goodness of fit of the residuals from the SARIMA and
GARCH models on the price of tea in Kenya.
3 To predict monthly tea prices and volatility in Kenya using the SARIMA and
GARCH models, respectively.

October 2022 Supervisors: Dr. Wycliffe C


Robin Ochieng Otieno SCT-325-002/2019 (Multimedia University of Kenya) 12 / 45
Significance of the Study

Significance of the study


• The impetus for doing this research stems from the issues confronting
the tea industry, notably the insecurity of growers’ revenues.
• The Tea Board of Kenya (TBK) will utilize the findings of this study
to make financial decisions regarding price risk analysis and to plan
for procurement needs because they will be able to estimate prices
and defend against the negative repercussions of price volatility.
• This will contribute to increased tea output, resulting in more tea
being exported and hence more money for tea-producing farmers and
an increase in the country’s foreign exchange profits and GDP.

October 2022 Supervisors: Dr. Wycliffe C


Robin Ochieng Otieno SCT-325-002/2019 (Multimedia University of Kenya) 13 / 45
Methodology

Introduction
• In this work, the SARIMA and GARCH models are used to model and
forecast tea prices and their volatility in Kenya using a time-series
method.
• The method entails identifying the price of tea pattern using a time
series plot, unit root testing for stationarity, determining the model
order, fitting and estimating the parameters of the appropriate
SARIMA and GARCH models, and lastly forecasting future prices.
• The data used in this study comprise the monthly price of tea in
Kenya over the period January 1, 1990 to November 1, 2021.

October 2022 Supervisors: Dr. Wycliffe C


Robin Ochieng Otieno SCT-325-002/2019 (Multimedia University of Kenya) 14 / 45
Methodology continued..

Test for stationarity and autocorrelation

• The Augmented Dickey-Fuller (ADF) test will be used to determine


stationarity; after stationarity is achieved, the autocorrelation and
partial autocorrelation plots will be inspected to determine their
correct structures.
• It will be conducted using the Ljung-Box Q-statistic test by Ljung and
Box (1978), which is defined as:
m
X rj2
Q = n(n + 2) (5)
n−j
j=1

October 2022 Supervisors: Dr. Wycliffe C


Robin Ochieng Otieno SCT-325-002/2019 (Multimedia University of Kenya) 15 / 45
Methodology continued..

Seasonal Auto-regressive Integrated Moving Average (SARIMA)


Model
• A multiplicative SARIMA model Xt can be denoted as SARIMA (p, d,
q) (P, D, Q)s; where (p, d, q) are the non-seasonal parameters, (P, D,
Q) are the seasonal parameters, and s is the seasonal period.
• The SARIMA model can be expressed as;

ΦP (B s )ϕ(B)∇D d s
s ∇ Xt = δ + ΘQ (B )θ(B)wt (6)
• The SARIMA coefficients are represented by Φp (B s ) and ΘQ (B s ) of
orders P and Q, while the ordinary and seasonal difference coefficients
are represented by ∇d = (1 − B) and ∇D s
s = (1−B ) Shumway et al.
(2000).
October 2022 Supervisors: Dr. Wycliffe C
Robin Ochieng Otieno SCT-325-002/2019 (Multimedia University of Kenya) 16 / 45
Methodology continued..

Generalized Auto-regressive Conditional


Heteroscedasticity(GARCH) model

• To represent the variance at time t, a GARCH model employs past


squared residuals and past variances.
• The GARCH (p, q) model combines the ARCH(q) model and
incorporates p lags of conditional variance in the model, where p is
known as the GARCH order.
• The combined model is called the GARCH(p,q) model given as:
p
X q
X
σt2 = α0 + 2
αi rt−i + 2
βj σt−j (7)
i=1 j=1

October 2022 Supervisors: Dr. Wycliffe C


Robin Ochieng Otieno SCT-325-002/2019 (Multimedia University of Kenya) 17 / 45
Methodology continued..

GARCH Model
• In this study we will use GARCH (1, 1) where the parameters p = 1
and q = 1 are given by:

σt2 = α0 + α1 rt−1
2 2
+ β1 σt−1 (8)

• Where σt2 is the conditional volatility, rt−1


2 are the previous months
2
squared returns, and σt−1 is the previous months volatility
• All coefficients α0 , α1 , and β1 are non negative this ensures that the
conditional volatility σt2 > 0 at all times

October 2022 Supervisors: Dr. Wycliffe C


Robin Ochieng Otieno SCT-325-002/2019 (Multimedia University of Kenya) 18 / 45
Methodology continued..

Model Selection
• The model selection will be based on the Akaike Information Criterion
(AIC) proposed by Akaike (1974).
• The Akaike Information Criterion approximates the quality of each
model given a set of data models, and the model with the least AIC
value is considered the best model.
• The formula for the Akaike Information Criterion (AIC) is:

AIC = 2K − 2 ln(L) (9)


• Where K is the number of estimated model parameters and L denotes
the maximum likelihood function for the model.

October 2022 Supervisors: Dr. Wycliffe C


Robin Ochieng Otieno SCT-325-002/2019 (Multimedia University of Kenya) 19 / 45
Methodology continued..

Model Diagnostics

• Following model selection and parameter estimation, the model


residuals will be used to perform model diagnostics, which will test
the independence and normality of the fitted model’s residuals.
• The Jaque Bera test will be used to determine normality, while the
Ljung Box test will be used to determine residual independence.
• The histogram density and the normal q-q plots are also used to
confirm if the standard residuals conform to a normal distribution.

October 2022 Supervisors: Dr. Wycliffe C


Robin Ochieng Otieno SCT-325-002/2019 (Multimedia University of Kenya) 20 / 45
Methodology continued..

Forecasting Performance Evaluation

• The Root Mean Square Error (RMSE) and Mean Absolute Error
(MAE) are used to evaluate the models’ predicting ability.
• The formula for the MAE and RMSE are given as:
n
1X
MAE = |Yt − Ft | (10)
n
t=1
v
u n
u1 X
RMSE = t (Yt − Ft )2 (11)
n
t=1

October 2022 Supervisors: Dr. Wycliffe C


Robin Ochieng Otieno SCT-325-002/2019 (Multimedia University of Kenya) 21 / 45
Results and Discussion.
Data
• To forecast the prices of tea and their volatility in Kenya, monthly
prices per kilogram, in Kenyan shillings, data for the period January
1, 1990, to November 1, 2021, obtained from FRED.
• The average price of tea over the years is Ksh. 245.80 with a
standard deviation of Ksh. 60.70. The series also shows positive
skewness of 0.591, meaning that the distribution has a long right tail.
Observations 381
Mean 245.8216
Median 236.6700
Maximum 403.0323
Minimum 143.3701
Std. Dev. 60.6912
Skewness 0.5909
Kurtosis 2.3701

October 2022 Supervisors: Dr. Wycliffe C


Robin Ochieng Otieno SCT-325-002/2019 (Multimedia University of Kenya) 22 / 45
Results and Discussion.
A Plot of the data
• The plot clearly shows that the data is non-stationary, with the series
trend growing with time. Seasonal and cyclic patterns may also be
seen throughout time.

October 2022 Supervisors: Dr. Wycliffe C


Robin Ochieng Otieno SCT-325-002/2019 (Multimedia University of Kenya) 23 / 45
Results and Discussion.

Test for Autocorrelation and Checking for Outliers


• Prior to the stationarity test, a lag plot was created to check for
autocorrelation and the presence of outliers. The plot shows that
there are no outliers in the data and that there is positive serial
auto-correlation. As a result, an auto-regressive model is preferable.

October 2022 Supervisors: Dr. Wycliffe C


Robin Ochieng Otieno SCT-325-002/2019 (Multimedia University of Kenya) 24 / 45
Results and Discussion.

Test for Stationarity


• The stationarity test was carried out in two ways: inferentially and
visually. It was done inferentially with the ADF test for stationarity
and visually with the ACF plot. The ADF test statistic is (−1.8424),
and the p-value is 0.3596, which is more than 0.05, confirming the
presence of a unit root and suggesting that the data is non-stationary.

Table: Results of the ADF test


Test Statistics -1.8424
P-Value 0.3596
Lags Used 10.0000
Critical Value(1%) -3.4481
Critical Value(5%) -2.8693
Critical Value(10%) -2.5709

October 2022 Supervisors: Dr. Wycliffe C


Robin Ochieng Otieno SCT-325-002/2019 (Multimedia University of Kenya) 25 / 45
Test for Stationarity.

AutoCorrelation Function Plot


• The ACF plot shows a gradually decaying ACF, implying that previous
values substantially impact future values of the monthly tea price in
the series. This is an example of an ACF plot for non-stationary data.

October 2022 Supervisors: Dr. Wycliffe C


Robin Ochieng Otieno SCT-325-002/2019 (Multimedia University of Kenya) 26 / 45
Attaining Stationarity.
Differencing the Series
• Because we can’t feed non-stationary data through our model, we
make the data stationary by taking the first difference in tea prices.
The plot of the stationary data is shown below, and it clearly shows a
consistent mean and variance over time.

October 2022 Supervisors: Dr. Wycliffe C


Robin Ochieng Otieno SCT-325-002/2019 (Multimedia University of Kenya) 27 / 45
The SARIMA Model

SARIMA Model Parameter Identification


• The AR and MA parameters are identified using the Extended
Autocorrelation Function (EACF) since the data is stationary. Based
on the EACF results, an ARMA(3,3) model was selected as the best
model to be fitted among other preliminary models.

October 2022 Supervisors: Dr. Wycliffe C


Robin Ochieng Otieno SCT-325-002/2019 (Multimedia University of Kenya) 28 / 45
The SARIMA Model
Model Selection
• In order to determine the optimal SARIMA model, several SARIMA
models of different parameter orders are fitted on the monthly price of
tea in Kenya. The optimal SARIMA model determined for fitting the
monthly price of tea in Kenya is SARIMA(3, 0, 3)(0, 1, 1)1 2 with an
AIC value of 3167.082, which is given below.

Table: The AIC values of the fitted models


Model AIC
SARIMA(1, 0, 1)(0, 1, 1)12 3168.693
SARIMA(1, 0, 0)(1, 1, 0)12 3274.834
SARIMA(2, 0, 1)(0, 1, 1)12 3169.513
SARIMA(2, 0, 3)(0, 1, 1)12 3172.886
SARIMA(1, 0, 2)(0, 1, 1)12 3168.906
SARIMA(3, 0, 3)(0, 1, 1)12 3167.082
SARIMA(1, 0, 1)(1, 1, 2)12 3172.675
October 2022 Supervisors: Dr. Wycliffe C
Robin Ochieng Otieno SCT-325-002/2019 (Multimedia University of Kenya) 29 / 45
The SARIMA Model

Coefficient Estimation
• The SARIMA(3, 0, 3)(0, 1, 1)12 model was used to estimate the eight
coefficients of the SARIMA model. It can be seen that the model
coefficients are all significant since they are all less than 0.05,
indicating that the SARIMA(3, 0, 3)(0, 1, 1)12 model is significant.
Coef P-Value Conclusion
AR(1) 2.130 0.000 Significant
AR(2) -1.913 0.000 Significant
AR(3) 0.753 0.000 Significant
MA(1) -1.025 0.000 Significant
MA(2) 0.497 0.000 Significant
MA(3) 0.205 0.000 Significant
MA.S(1) -0.971 0.000 Significant
Sigma 267.178 0.000 Significant

October 2022 Supervisors: Dr. Wycliffe C


Robin Ochieng Otieno SCT-325-002/2019 (Multimedia University of Kenya) 30 / 45
The SARIMA Model
Residual Diagnostics
• The histogram shows that the standard residuals have a normal
distribution. The normal probability plot showed that the data had a
normal distribution, and the correlogram showed no substantial lag,
indicating that the model was adequate.

October 2022 Supervisors: Dr. Wycliffe C


Robin Ochieng Otieno SCT-325-002/2019 (Multimedia University of Kenya) 31 / 45
The SARIMA Model

Forecasting Monthly tea prices to 2024


• The SARIMA(3, 0, 3)(0, 1, 1)12 model was used to forecast future
monthly prices of tea up to June 2024.

October 2022 Supervisors: Dr. Wycliffe C


Robin Ochieng Otieno SCT-325-002/2019 (Multimedia University of Kenya) 32 / 45
The SARIMA Model
Tea Price forecasts from December 2022 to June 2024
Date Prediced price
2022-12-01 259.24
2023-01-01 259.33
2023-02-01 254.25
2023-03-01 250.32
2023-04-01 242.72
2023-05-01 241.38
2023-06-01 241.79
2023-07-01 247.01
2023-08-01 251.47
2023-09-01 259.84
2023-10-01 258.23
2023-11-01 253.23
2023-12-01 252.68
2024-01-01 257.70
2024-02-01 258.01
2024-03-01 249.53
2024-04-01 241.71
2024-05-01 240.36
2024-06-01 240.69
October 2022 Supervisors: Dr. Wycliffe C
Robin Ochieng Otieno SCT-325-002/2019 (Multimedia University of Kenya) 33 / 45
The SARIMA Model

Evaluating the SARIMA Model


• The SARIMA model’s prediction capacity was evaluated using the
Root Mean Squared Error and Mean Absolute Error indicated below.
We may thus infer that the SARIMA(3, 0, 3)(0, 1, 1)12 model
performed well in terms of monthly tea prices forecasting.

Table: Forecasting performance of the SARIMA Model


Model RMSE MAE MSE
SARIMA(3, 0, 3)(0, 1, 1)12 17.90 11.90 320.66

October 2022 Supervisors: Dr. Wycliffe C


Robin Ochieng Otieno SCT-325-002/2019 (Multimedia University of Kenya) 34 / 45
Volatility Modeling

Plot of the Returns


• It can be observed from the plot that the returns are stationary; that
is, they have a constant mean and variance over time

October 2022 Supervisors: Dr. Wycliffe C


Robin Ochieng Otieno SCT-325-002/2019 (Multimedia University of Kenya) 35 / 45
Volatility Modeling

Descriptive Statistics of the Returns


• To further comprehend our return series, we computed descriptive
statistics for tea’s monthly returns. The lowest return was -21.038,
the highest was 23.956, the mean was 0.346, and the standard
deviation was 6.974. With a skewness of 0.224, the results are
considerably skewed.

Table: Descriptive statistics of the monthly Returns


Observations Minimum Mean Std. Dev Maximum Skewness Kurtosis
381 −21.038 0.346 6.974 23.956 0.224 3.617

October 2022 Supervisors: Dr. Wycliffe C


Robin Ochieng Otieno SCT-325-002/2019 (Multimedia University of Kenya) 36 / 45
Volatility Modeling

Test for Heteroscedasticity


• The Heteroscedasticity test is used to determine the presence of arch
effects in the squared residuals. It is calculated using the Lagrange
Multiplier. We reject the null hypothesis since the ARCH-LM test
statistic is 3.875 with a P-Value of 0.0490, and we conclude that
there are significant Arch effects from January 1, 1990 to November
1, 2021. As a consequence, we can determine which GARCH model is
optimal.
Returns Chi-square P-Value
Tea 3.875 0.0490

October 2022 Supervisors: Dr. Wycliffe C


Robin Ochieng Otieno SCT-325-002/2019 (Multimedia University of Kenya) 37 / 45
Volatility Modeling - The GARCH Model

Model Selection
• Different GARCH(p,q) models are simulated on the price of tea data,
and the results show that GARCH(1,1) is the best model for fitting
monthly tea price data, with an AIC of 2542.47 and a BIC of 2558.24.
As a result, the GARCH(1,1) model coefficients are estimated for the
monthly price of tea in Kenya.
Model AIC BIC
GARCH(0, 1) 2563.33 2575.15
GARCH(0, 2) 2545.57 2561.34
GARCH(1,1) 2542.47 2558.24
GARCH(1, 2) 2543.97 2563.68
GARCH(2, 1) 2544.47 2564.19
GARCH(2, 2) 2543.79 2567.45

October 2022 Supervisors: Dr. Wycliffe C


Robin Ochieng Otieno SCT-325-002/2019 (Multimedia University of Kenya) 38 / 45
Volatility Modeling - The GARCH Model
Coefficient Estimation
• The GARCH(1,1) model was used for coefficient estimation, and the
p-values are all smaller than α = 0.05. As a result, we conclude that
all of the GARCH(1,1) model’s coefficient are relevant and the
GARCH(1,1) model is the optimal model.
Coefficient P-Value Conclusion
α0 2.9566 9.135e − 02 Significant
α1 0.1527 1.022e − 02 Significant
β1 0.7952 5.531e − 34 Significant
The volatility model GARCH(1,1) for the Monthly price of tea can
thus be expressed as:

σt2 = 2.9566 + 0.1527rt−1


2 2
+ 0.7952σt−1 (12)
October 2022 Supervisors: Dr. Wycliffe C
Robin Ochieng Otieno SCT-325-002/2019 (Multimedia University of Kenya) 39 / 45
Volatility Modeling - The GARCH Model.
Residual Diagnostics - Probability Plot
• The GARCH(1,1) model diagnostics were conducted on the
standardized residuals, which are supposed to follow a normal
distribution. The normality of the residuals is checked using the
probability plot and the histogram density plot.

October 2022 Supervisors: Dr. Wycliffe C


Robin Ochieng Otieno SCT-325-002/2019 (Multimedia University of Kenya) 40 / 45
Volatility Modeling - The GARCH Model.
Residual Diagnostics - Histogram Density Plot
• The histogram density plot further demonstrates that the normalized
residuals follow a normal distribution since the histogram is
symmetrical in shape. The GARCH(1,1) model residuals are therefore
proved to be significant.

October 2022 Supervisors: Dr. Wycliffe C


Robin Ochieng Otieno SCT-325-002/2019 (Multimedia University of Kenya) 41 / 45
Volatility Modeling - The GARCH Model.
Forecasting Volatility
• The train set is used to fit the GARCH(1) model and forecast for the
next 36 months. The GARCH(1,1) model forecast data pattern and
the actual data pattern are practically identical. As a result, we
continued to evaluate the forecasts for the GARCH(1,1) model.

October 2022 Supervisors: Dr. Wycliffe C


Robin Ochieng Otieno SCT-325-002/2019 (Multimedia University of Kenya) 42 / 45
Volatility Modeling - The GARCH Model.

Evaluation of Forecasts
• Evaluation of the forecast of the GARCH(1,1) model is conducted
using the Mean Absolute Error (MAE) and the Root Mean Squared
Error(RMSE). The RMSE (5.903) and the MAE (5.089) values are
significant. Thus we conclude that GARCH(1,1) model performed
well in forecasting volatility.

Table: Forecasting performance of the GARCH(1,1) Model

Model RMSE MAE


GARCH(1, 1) 5.903 5.089

October 2022 Supervisors: Dr. Wycliffe C


Robin Ochieng Otieno SCT-325-002/2019 (Multimedia University of Kenya) 43 / 45
Conclusion and Recommendation.

• The SARIMA(3, 0, 3)(0, 1, 1)12 model was identified as the best


model for predicting tea prices in Kenya using the EACF plot. The
SARIMA model’s eight parameters were all found to be significant,
with p-values less than 0.05. The GARCH(1, 1) model was identified
as the best GARCH model for forecasting tea price volatility.
• The long-run volatility persistence level determined using the
GARCH(1, 1) model parameters α1 + β1 = 0.9479, indicating lengthy
durations of future volatility shock persistence.
• For further research, I would advocate using the SARIMA model
residuals to estimate the parameters of the GARCH(1,1) model,
resulting in a SARIMA-GARCH model that may be used to predict
volatility.

October 2022 Supervisors: Dr. Wycliffe C


Robin Ochieng Otieno SCT-325-002/2019 (Multimedia University of Kenya) 44 / 45
References.
Akaike, H. (1974). A new look at the statistical model identification. IEEE
transactions on automatic control, 19(6), 716–723.
Chinomona, A. (2009). Time series modeling with application to south
african inflation data (Doctoral dissertation).
Fwaga, S. O., Orwa, G., & Athiany, H. (2017). Modelling rates of inflation
in kenya: An application of garch and egarch models.
Mathematical Theory and Modelling, 7(5), 75–83.
Gesimba, R., Langat, M., Liu, G., & Wolukau, J. (2005). The tea industry
in kenya; the challenges and positive developments. Journal of
Applied Sciences, 5(2), 334–336.
KIPPRA. (2020). Fluctuations in market earnings for tea in kenya: What
could be the cause and remedy?
https://kippra.or.ke/fluctuations-in-market-earnings-for-tea-in-
kenya-what-could-be-the-cause-and-remedy/
Ljung, G. M., & Box, G. E. (1978). On a measure ofOctober
lack2022
of Supervisors:
fit in time Dr. Wycliffe C
Robin Ochieng Otieno SCT-325-002/2019 (Multimedia University of Kenya) 45 / 45

You might also like