NATIONAL ECONOMICS UNIVERSITY

Faculty of Mathematical Economics

-------***-------

ASSIGNMENT
ECONOMETRICS 2

Topic: Stock price and Return volatility

forecasting using Time Series Model

Full name: Nguyen Ngoc Khoa

Class: Actuary 63
Student’s ID: 11212889

Hanoi, 2023
Stock price and Return volatility forecasting using Time Series Model

CONTENT

1. Introduction.......................................................................................3
2. Forecasting by ARIMA and ARCH-GARCH................................3
2.1. Methodology.....................................................................................3
2.2. ARIMA modelling results................................................................8
2.3 ARCH-GARCH modelling results.................................................17
3 Conclusion........................................................................................19
4 References........................................................................................19

2
Nguyễn Ngọc Khoa– Actuary 63
Stock price and Return volatility forecasting using Time Series Model

1. Introduction

These days, stock market has become an attractive and interesting investment channel
around the world for not only investment institutions but also individual investors, and
Vietnam is not lying out of the trend. Vietnam stock’s market is established in 2000s and
now it has three popular stock exchanges: HOSE, HNX and UPCOM. During and after
the COVID-19 pandemics, the number of people joining stock market increases
significantly, but most people invest emotionally and base on the short-term information.
Besides that, there are also some unknown factors affecting the stock market. Then, the
stock price always fluctuates tremendously. Therefore, forecasting the trend of stock
price is necessary for individuals and organizations to build a suitable investment
strategy. I suggest the ARIMA model and ARCH-GARCH model to forecast for the
stock price and volatility of return in the short term based on historical data.

I will be using the stock BCM (Becamex IDC Corp) listed on HOSE to analyze
forecast future values. BCM, another name of Becamex Investment & Industrial
Development, after more than 40 years of operating, Becamex IDC has become the
leading brand in the fields of investment and construction of industrial zone, residential
area, urban area and transportation infrastructure. It is also renowned for sponsoring a
successful football team playing in VLeague. Becamex IDC was founded in 1976 under
the name of Bến Cát General Trade Company (Becamex), in 1999 the company changed
its name to Trade – Investment and Development Company (BECAMEX Corp). For the
past 3 decades, the Becamex is the key driver behind Binh Duong’s remarkable
industrialization during the past 3 decades. Yet, it remains our shared commitment to
accelerate the economic transformation of Vietnam, moving up the global value chain,
embracing the Industry 4.0 revolution and beyond.

2. Forecasting by ARIMA and ARCH-GARCH

2.1. Methodology
2.1.1 Data collection

The study takes data from website investing.com.vn, which is a trustworthy source
providing information about the Vietnamese Stock market. The collected data is the daily
closing share price of Becamex IDC Corp (BCM), from 1st January 2022 to 31st July
2023. Thus, there is a total of 391 observations. This study will use the last 10
observations as validation set and the previous series as training set.

3
Nguyễn Ngọc Khoa– Actuary 63
Stock price and Return volatility forecasting using Time Series Model

2.1.2 ARIMA Model

Autoregressive Integrated Moving Average (ARIMA) is an efficient quantitative

forecasting model for time series. The Autoregressive (AR) is one where the current
value of a variable yt depends upon only the values that the variable took in previous
periods. The Moving Average (MA) is simply a linear combination of white noise
processes, so that the yt depends on the current and previous values of a white noise
disturbance term (Brooks C. (2014)). Integrated order named as “I” is the order of
difference of the time series. The ARIMA model is exercised by Box-Jenkins method
(Box, G. E. P. and Jenkins, G. M. (1976)) including 4 steps: Identification, Estimation,
Diagnostic Checking and Forecasting.

The first step: Identification.

This includes checking stationarity and determining order of ARIMA. The Dickey-
Fuller test will be used to check for stationarity. The null hypothesis is that the series is
unit root and non-stationary. If the DF statistic is larger than the critical value, then we
reject the null hypothesis, the series is stationary. Next, I will plot the PACF and ACF of
the stationary series to determine the order. The order of AR is order of the series having
partial correlation while the order of MA is the order of the series having auto correlation.

The second step: Estimation.

After choosing order for ARIMA, I will estimate the intercept and coefficients of this
model. With the criteria of significant coefficients and AIC criteria, I will choose good
models to go on the next step.

The third step: Diagnostic Checking.

In this part, I will check the stationarity by using inverse roots and unit circle. If all
the inverse roots lie strictly inside the circle, then the series is stationary. Furthermore, I
check that whether the residual series is white noise with mean equal zero, constant
variance, no serial correlation and normal distribution.

The fourth step: Forecasting.

I will use the ARIMA models after diagnostic checking to forecast for the validation
data. The model with smaller forecast error will be chosen to forecast for the future.

4
Nguyễn Ngọc Khoa– Actuary 63
Stock price and Return volatility forecasting using Time Series Model

Figure 1. Box-Jenkins Methods illustrates the process

The general model of ARIMA ( p , d ,q ) is:

Y t =μ+(ϕ 1 Y t −1+⋯+ ϕ p Y t − p)+(ε t +θ1 ε t −1 +⋯+θ q ε t −q)

In which:

{
Y t is the stationary I (d ) series
μ is theintercept
ε t is the shock of Y t at timet
p isthe order of AR (p)
ϕ p is the coefficient of AR( p)
q is the order of MA (q )
θ q is the coefficient of MA (q)

2.1.3 ARCH-GARCH Model

The autoregressive changing heteroscedasticity model of order p, ARCH(p), is models

of volatility developed by Engle (1982), providing a framework for the analysis and
development of time series. This model assumes that the variance at time t depends on
the shock in the previous stages. However, one of the drawbacks of the ARCH model is
that it looks like a MA model rather than AR model (Engle (1995)). So, a new idea is
proposed that lagged variables should be added into the variance equation according to
autoregressive form. This model is called “Generalized Autoregressive Conditional
Heteroskedasticity” (GARCH), developed independently by Bollerslev (1986) and Taylor

5
Nguyễn Ngọc Khoa– Actuary 63
Stock price and Return volatility forecasting using Time Series Model

(1986). The GARCH model is also used to forecast for volatility, allowing the
conditional variance to be dependent upon previous own lags.

The model of ARCH( p):

2 2 2
σ t =w+ γ 1 ε t −1+ ⋯+ γ p ε t− p + v t

The model of GARCH( p , q):

2 2 2 2 2
σ t =w+ δ 1 σ t −1+⋯+ δ p σ t− p +γ 1 ε t−1 +⋯+ γ q ε t −q + vt

In which:

{
2
σ t is the conditional variance of ε t
ε t is the shock at timet
w is long−run volatility parameter ensure the variance is positive ¿
p , q is theorder of GARCH ( p , q)
γ i ≥0 , δ i ≥ 0 should be held
¿

The volatility is stationary if ∑ δ i + ∑ γ i <1

Thus, the unconditional variance is:

2 w
σ =
1−¿ ¿

2.1.4 Variables and formulas

Table 1. Variables and formulas

Variables Formulas Meaning

Pt The price of BCM at time t
∆t ∆ t=P t−Pt −1 The 1st difference of price of
BCM at time t
lr t Pt The log-return rate of BCM
r t =ln ( )
Pt −1 at time t

6
Nguyễn Ngọc Khoa– Actuary 63
Stock price and Return volatility forecasting using Time Series Model

2.1.5 Plot and descriptive statistics

Figure 2. Plot of price

Table 2. Descriptive statistics of price

Price
Mean 86896.68
Median 87000.00
Minimum 64500.00
Maximum 116300.00
Range 51800.00
Standard Error 560.13
Standard Deviation 11075.87
Sample Variance 122674937.64
Kurtosis -0.23
Skewness 0.33
Observations 391

The table shows the descriptive measures for the investigated price series. The table
indicates that the mean price over the period is 86896.68 and standard deviation is

7
Nguyễn Ngọc Khoa– Actuary 63
Stock price and Return volatility forecasting using Time Series Model

11075.17. Meanwhile, the plot shows that there is a upward trend in the stock price. This
information can give a first glance that the series is trend stationary. the price fluctuates
significantly over the period shown. To have a further analysis about the stationarity and
volatility of the price, the study will apply the ARIMA model and ARCH-GARCH model
in the next step.

2.2. ARIMA modelling results

2.2.1 Dickey-Fuller test

The study uses the Dickey Fuller test to test the stationarity of the price series. The
result is presented in the table below:

Table 3. Dickey Fuller test for price series

Critical value at 1%
Test DF test statistics (τ¿ ¿ stat )¿
(τ¿ ¿ 0.01)¿
Dickey-Fuller test with trend
Price series −3.0136 −3.98

It is shown that τ stat is smaller than τ 0.01, then the study does not reject the hypothesis
that the series is unit root at level of significance of 1%. Thus, the price series is non-
stationary, and it is necessary to transform this series into the stationary form to have a
better forecasting. This study will take the 1st difference to the original series and log-
return series in ARIMA model applications.

The table below shows the results of Dickey Fuller test for 1st difference series and
log-return series.

Table 4. Dickey Fuller test for difference and log-return series

Critical value at 1%
Test DF test statistics (τ¿ ¿ stat )¿
(τ¿ ¿ 0.01)¿
Dickey-Fuller test without drift
Difference series −15.893 −2.58
Log-return series −18.9238 −2.58

It is show that τ stat is greater than τ 0.01 for both difference series and log-return series,
then both series are not unit root at level of significance of 1%. Thus, difference series
and log-return series can be used to apply ARIMA model.

8
Nguyễn Ngọc Khoa– Actuary 63
Stock price and Return volatility forecasting using Time Series Model

2.2.2 ACF and PACF

The two figures below show the plot of ACF and PACF of the 1st difference series.

Figure 3. PACF plot of 1st difference Figure 4. ACF plot of 1st difference

In the PACF plot, the series having partial correlation is 2, 6, 7 and 10. Meanwhile, in
the ACF plot, the series has autocorrelation at order 2, 6, 7, 8, 10 and order 15 is the
furthest order having autocorrelation. Thus, it is indicated that the 1 st difference series
uses the order 4 for AR and order 7 for MA in ARIMA model.

The two figures below show the plot of ACF and PACF of the return rate series.

Figure 5. PACF plot return rate Figure 6. ACF plot of return rate

9
Nguyễn Ngọc Khoa– Actuary 63
Stock price and Return volatility forecasting using Time Series Model

In the PACF plot, the series has partial autocorrelation at order 2, 7 and 7 is also the
furthest order having partial autocorrelation. Meanwhile, in the ACF plot, the series has
autocorrelation at order 2, 6, 7, 8, 10 and order 15 is the furthest order having
autocorrelation.

2.2.3 ARIMA Models

Table 5. ARIMA models result for 1st difference series

ARIMA (5,1,2) (6,1,2) (7,1,2) (7,1,2) fixed (2,1,2)

ϕ1 0.531*** -0.1049 0.4926*** 0.5456*** 0.6593***
ϕ2 -0.9371*** -0.703*** -0.7741*** -0.8354*** -0.9664*
ϕ3 0.0664 -0.0329 0.0416 - -
ϕ4 -0.1246* -0.125** -0.1301** -0.1259*** -
ϕ5 0.0912* 0.0326 0.0205 - -
ϕ6 - 0.1186** 0.0426 - -
ϕ7 - - -0.1438*** -0.1283*** -
θ1 -0.513*** 0.132 -0.4728*** -0.545*** -0.6338***
θ2 0.8175*** 0.5894*** 0.6582*** 0.7379*** 0.8998***
AIC 6764,9 6767.28 6763.81 6758.66 6761.86
[*], [**], [***]: significant at 10%, 5%, 1%

Comparing five models, we can firstly reject models ARIMA (5,1,2), ARIMA (6,1,2)
and ARIMA (7,1,2) as the AR and MA coefficients are not all significant even though the
intercepts of these three models are significant at 10%. Thus, if we use these models to
forecast, the results can be biased. Next, we consider the remaining models. Both
ARIMA (7,1,2)-fixed and ARIMA (2,1,2) have significant intercepts and coefficients at
level of significance of 10%. Since the AIC of ARIMA (7,1,2)-fixed is the smallest
therefore ARIMA (7,1,2)-fixed is the most suitable model of the above. However, I also
put ARIMA (2,1,2) to the diagnostic checking part because of the recommendation of the
RStudio.

10
Nguyễn Ngọc Khoa– Actuary 63
Stock price and Return volatility forecasting using Time Series Model

Table 6. ARIMA models result for log-return rate series

ARIMA (2,0,2) (2,0,2)-no (7,0,2)- no (7,0,2) fixed (7,0,6)

intercept intercept fixed
Intercept 0.0005 - -
ϕ1 0.6604*** 0.6604*** 0.5553*** 0.5557*** 0.3406***
ϕ2 -0.9717*** -0.9725*** -0.8297*** -0.8311*** -0.1154*
ϕ3 - - 0.039 - -1.1078**
ϕ4 - - -0.1309** -0.1116*** -
ϕ5 - - 0.0301 - 0.2357***
ϕ6 - - 0.0032 - -0.8665***
ϕ7 - - -0.1242* -0.1272*** -
θ1 -0.6357*** -0.6359*** -0.5346*** -0.5483*** -0.2515***
θ2 0.9062*** 0.9077*** 0.7188*** 0.735*** -0.0064
θ3 - - - - 1.1452***
θ4 - - - - 0.108***
θ5 - - - - -0.3089***
θ6 - - - - 0.9076***
AIC -1783.16 -1784.96 -1781.11 -1786.75 -1782.13
[*], [**], [***]: significant at 10%, 5%, 1%

Comparing these five models, we can firstly reject ARIMA (2,0,2) due to the
insignificant intercept at 10%. Similarly, not all the coefficients in ARIMA (7,0,2)- no
intercept and ARIMA (7,0,6) fixed are significant at level of significance of 10%,
therefore; we reject these three models. Next, we consider the remaining models. Both
ARIMA (2,0,2)-no intercept and ARIMA (7,0,6) fixed have significant intercepts and
coefficients at level of significance of 10%, so the AIC will be put on the table. As the
AIC of ARIMA (7,0,2) fixed is the smaller one, therefore; we can conclude ARIMA
(7,0,2) fixed is the most suitable model of the above but the ARIMA (2,0,2) no intercept
still meets the requirements and put it to the checking residual pharse

11
Nguyễn Ngọc Khoa– Actuary 63
Stock price and Return volatility forecasting using Time Series Model

2.2.4 Diagnostic checking

The first diagnostic checking is stationary checking by inverse roots and unit circle. If
the series is stationary, then all the inverse roots should lie strictly within the AR unit
circle and MA unit circle

Figure 7. Unit circle of ARIMA (2,1,2) Figure 8. Unit circle of ARIMA (7,1,2) fixed

Figure 9. Unit circle of ARIMA (2,0,2) Figure 10. Unit circle of ARIMA (7,0,2) fixed

Figure 7 and 8 shows that all inverse roots

lie inside the unit circle, so we can strictly
conclude that the 1st difference series is
stationary. Like the 1st difference series, all
inverse roots in Figure 9, 10 lie inside the
unit circle, so we can strictly conclude that the log-return series is stationary.

The second diagnostic checking is testing for residual series.

Table 7. Residual series testing

ARIMA (2,1,2) (7,1,2) fixed (2,0,2) no (7,0,2) fixed

12
Nguyễn Ngọc Khoa– Actuary 63
Stock price and Return volatility forecasting using Time Series Model

intercept
Degree of 6 3 6 3
freedom
P-value 0.07791 0.071 0.1121 0.03441
(Residual test)

Figure 11. Residual series of ARIMA (2,1,2)

Figure 12. Residual series of ARIMA (7,1,2) fixed

Figure 13. Residual series of ARIMA (2,0,2) no intercept

Figure 14. Residual series of ARIMA (7,0,2) fixed

13
Nguyễn Ngọc Khoa– Actuary 63
Stock price and Return volatility forecasting using Time Series Model

For ARIMA (7,0,2) fixed, we reject hypothesis since the P-value is 3.441%, lower
than the significance level of 5%.For the rest model ARIMA (2,1,2), ARIMA (7,1,2)
fixed and ARIMA (2,0,2) as the P-value is higher than 5%. Thus, residual series of the
rest models are white noises. After two diagnostic tests, it is obvious that we can apply
ARIMA (2,1,2), ARIMA (7,1,2) fixed and ARIMA (2,0,2) models to forecast for the
validation data.

2.2.5 Forecasting for validation set

To find the best model among ARIMA (2,1,2), ARIMA (7,1,2) fixed, ARIMA (2,0,2)
no intercept, I will apply these models to forecast for the validation set. The model with
smaller forecast error will be the best model and be chosen to forecast for future stock
price.

Table 8. Forecasting results for 1st difference and log-return

Date 1st difference (2,1,2) 1st difference (7,1,2) Log return (2,0,2)
fixed
18/07/2023 50.1620 143.45 0.001786218
19/07/2023 -76.6798 -158.27 -0.001194465
20/07/2023 -248.2627 -257.76 -0.00252595
21/07/2023 -220.7552 143.4 -0.000506511
24/07/2023 -89.3370 237.02 0.00212202
25/07/2023 -44.4273 42.28 0.001893978
26/07/2023 -102.6040 -27.03 -0.000812905
27/07/2023 -150.1752 -86.53 -0.002378765
28/07/2023 -129.9253 -34.16 -0.000780382
31/07/2023 -89.6467 81.41 0.001798015

Table 9. Forecasting results for stock price using ARIMA (2,1,2).

Date (2,1,2) Actual price Error

18/07/2023 80142.34 79800 0.429%

14
Nguyễn Ngọc Khoa– Actuary 63
Stock price and Return volatility forecasting using Time Series Model

19/07/2023 80053.8 80100 -0.057%

20/07/2023 79857.87 80300 -0.551%
21/07/2023 79814.24 79600 0.269%
24/07/2023 79974.83 79200 0.978%
25/07/2023 80122.87 78000 2.722%
26/07/2023 80065.29 78400 2.124%
27/07/2023 79884.25 78500 1.763%
28/07/2023 79820.54 78000 2.334%
31/07/2023 79953.49 81000 -1.291%

Table 10. Forecasting results for stock price using ARIMA (7,1,2) fixed.

Date (7,1,2) fixed Actual price Error

18/07/2023 80143.45 79800 0.43%
19/07/2023 79985.18 80100 -0.143%
20/07/2023 79727.42 80300 -0.713%
21/07/2023 79870.82 79600 0.34%
24/07/2023 80107.84 79200 1.146%
25/07/2023 80150.12 78000 2.756%
26/07/2023 80123.09 78400 2.198%
27/07/2023 80036.56 78500 1.957%
28/07/2023 80002.4 78000 2.567%
31/07/2023 80083.81 81000 -1.131%

Table 11. Forecasting results for stock price using ARIMA (2,0,2) no intercept.

Date (2,0,2) no intercept Actual price Error

18/07/2023 80143.02 79800 0.43%
19/07/2023 80047.35 80100 -0.066%
20/07/2023 79845.41 80300 -0.566%
21/07/2023 79804.98 79600 0.258%
24/07/2023 79974.51 79200 0.978%
25/07/2023 80126.12 78000 2.725%
26/07/2023 80061.01 78400 2.118%
27/07/2023 79870.79 78500 1.746%
28/07/2023 79808.49 78000 2.318%
31/07/2023 79952.11 81000 -1.293%

From Table 9, we can see that the further the lapse of time, the higher difference
between actual data and the forecasting results. However, the forecasting results have less
than 5% error, within allowable limits. According to this result, the stock price of BCM
tends to decrease gradually soon. To have better comparison between two models and
forecast evaluation, I will calculate the forecast error through three criterias: RMSE,
MAE, MAPE.

15
Nguyễn Ngọc Khoa– Actuary 63
Stock price and Return volatility forecasting using Time Series Model

Table 12. Forecast error

Criteria ARIMA (2,1,2) ARIMA (7,1,2) fixed ARIMA (2,0,2) no

intercept
RMSE 1203.997 1271.164 1203.996
MAE 985.920 1053.787 984.405
MAPE 1.527398 0.020564 0.01518

From Table 12, the RMSE, MAE and MAPE of ARIMA (2,0,2) no intercept is the
smallest of all three ARIMA model. Thus, we can conclude that model ARIMA (2,0,2)
no intercept used for log-return series which is the best model to forecast. The equation
of ARIMA (2,0,2) no intercept is:
r t =0.6604 r t−1−0.9725 r t −2+ ε t −0.6359 ε t−1+ 0.9077 ε t −2

2.2.6 Forecasting for future

In this part, to forecast for the stock price for the next 10 days from 31/07/2023, I use
model ARIMA (2,0,2) no intercept for log-return series with full data from 1/1/2022 to
31/07/2023 (a total of 391 observations). The forecasting method is mixed. The following
table is the stock price forecasting of VNM for the next 10 days.

Table 13. Future forecasting

Date Forecast price Actual price Error

1/8/2023 79996.01 79000 1.26%
2/8/2023 80059.33 77900 2.77%
3/8/2023 80105.08 77400 3.49%
4/8/2023 80073.70 77200 3.72%
7/8/2023 80008.55 76700 4.31%
8/8/2023 79996.00 76700 4.30%
9/8/2023 80051.01 76100 5.19%
10/8/2023 80099.59 74000 8.24%
11/8/2023 80078.17 74100 8.07%
14/8/2023 80016.83 73500 8.87%

From Table 12, the forecast stock price of BCM continued to experience a slight
decrease in the beginning of August. This trend is also witnessed in the actual price but
more significant. However, the forecasting results have exceeded 5% error, within the
allowable limits.

16
Nguyễn Ngọc Khoa– Actuary 63
Stock price and Return volatility forecasting using Time Series Model

2.3 ARCH-GARCH modelling results

2.3.1 Models

Chi-squared 29.377

Degree of freedoms 1

P-value 5.958e-08

As the log-return series is stationary around zero, I will forecast for the volatility of
the log-return of BCM stock prices. I use the ARCH test from order 1 for the log-return
series, and it is indicated that the P-value of the ARCH test for order 1 to order 98 are
smaller than the significant level of 5% .

Therefore, the series has ARCH effects to order 98. However, in order to have a better
forecast for volatility, the GARCH test for the series will be added. And the available
models for GARCH test are presented in the table below.

Table 14. ARCH-GARCH Models

GARCH (0,1) (0,2) (0,3) (0,4) (1,1) (1,2)

0.000228 0.00007988 0.00006069 0.00004067 0.000001292 0.000002573
w
[***] [***] [***] [***] [**] [**]
0.7384 0.7416 0.6749 0.5483 0.09465 0.2376
γ1
[***] [***] [***] [***] [***] [**]
0.6927 0518 0.2778
γ2 - - 5.255e-09
[***] [***] [***]
0.2098 0.1581
γ3 - - - -
[**] [*]
0.337
γ4 - - - - -
[***]
0.9017 0.8138
δ1 - - - -
[***] [***]

17
Nguyễn Ngọc Khoa– Actuary 63
 Stock price and Return volatility forecasting using Time Series Model

δ2 - - - - - -

As the table presents, we can see that ARCH (4) model has an insignificant
coefficient at the level of significance of 5%. And when I increase the order of ARCH
model, there are still some insignificant coefficients. So the ARCH (3) model is the best
one to forecast among the ARCH model. However, GARCH is a better model to forecast
than ARCH (Bollerslev (1986) and Taylor (1986)) as lagged variances are added. We can
see that all the intercept and coefficients of GARCH (1,1) are significant at 5%, while
GARCH (1,2) has one coefficient that is insignificant at 5%. To conclude, the GARCH
(1,1) is the best model to forecast for the volatility of BCM’s log-return.

Table 15. GARCH(1,1) test

Jarque Bera Test Box-Ljung Test

P-value 2.22e-16 0.05664

X-squared 72.339 0.36331

Degree of freedoms 2 1

In the context that GARCH (1,1) is also a popular model in academic research of votality
(Hansen & Lunde, 2005), the study would choose this model for analysis & prediction.
The model of GARCH (1,1) for residuals of Logreturn ARIMA (2,0,2) no intercept series
is:
2 2 2
σ t =0.000001292+0.09465 σ t −1+ 0.9017 ε t −1+ v t

2.3.2 Unconditional variance

The estimated unconditional variance of GARCH (1,1) is:

2 0.000001292
σ = =0.00035
1−0.09465−0.9017

18
Nguyễn Ngọc Khoa– Actuary 63
Stock price and Return volatility forecasting using Time Series Model

2.3.3 Forecast for volatility

In this part, to forecast for the volatility of the log-return series for the next day
1/8/2023, I use the above model GARCH (1,1) with fixed forecasting method. The
following table is the volatility forecasting for the next 10 days.

Table 15. Forecasting for volatility

Date Forecast volatility (σ 2t )

1/8/2023 0.00005228
2/8/2023 0.00005684
3/8/2023 0.00005825
4/8/2023 0.00005869
7/8/2023 0.00005882
8/8/2023 0.00005887
9/8/2023 0.00005888
10/8/2023 0.00005888
11/8/2023 0.00005888
14/8/2023 0.00005888

3 Conclusion and limitation

The forecast results in August show that the forecast price is approximately the actual
value with the error smaller than the allowable limit 10%. This proves that the reliability
of the ARIMA (2,0,2) no intercept for log-return series is high. Meanwhile, the
GARCH (1,1) reflects the volatility of log-return rate for 10 days, helps to estimate the
risk of BCM and how the return fluctuates in the future.

However, there are some limitations in forecasting by ARIMA and ARCH-GARCH.

In some trading sessions, the impact of major external factors such as high leverage trade,
emotional investments, information about major changes, or even some unpredictable
events will make the forecast error higher. Therefore, the results of the model are still
only more references. However, it can be said that the ARIMA model and ARCH-
GARCH model are good to forecast in the short term

19
Nguyễn Ngọc Khoa– Actuary 63
Stock price and Return volatility forecasting using Time Series Model

4 References
1. Brooks C. (2014) Introductory Econometrics for Finance, 3E.

2. Box, G. E. P. and Jenkins, G. M. (1976) Time Series Analysis: Forecasting and

Control, 2nd edn, Holden-Day, San Francisco.

3. Engle, R. F. and Kroner, K. F. (1995) Multivariate Simultaneous Generalised GARCH,

Econometric Theory 11, 122–50.

4. Bollerslev, T. (1986) Generalised Autoregressive Conditional Heteroskedasticity,

Journal of Econometrics 31, 307–27

5. Taylor, S. J. (1986) Forecasting the Volatility of Currency Exchange Rates,

International Journal of Forecasting 3, 159–70.

6. Denzil Watson, Antony Head (2016) Corporate Finance: Principles and Practice

8th Edition.

7. Markowitz (1959) Portfolio Selection: Efficient Diversification of Investments.

20
Nguyễn Ngọc Khoa– Actuary 63