Time-Series Forecasting of WTI Crude and Brent Crude Oil Prices 1

Oil-Price Modelling & Forecasting Based on

Univariate and Multivariate

Time-Series Models:

WTI Crude Oil & Brent Crude Oil Prices (1987-2020)

BY: Ankita, Aman, Nisha, Tanya, Purmina, Nimisha

Time-Series Forecasting of WTI Crude and Brent Crude Oil Prices

INTRODUCTION

Oil is one of the most important commodities in day to day life and its price and volatility has huge

impact in economic spheres. History is replete with examples in which nations have fought for this

precious Black Gold and will continue to do so until this commodity plays the central part of

human existence. Being, one of the fastest depleting non-renewable source of energy, it has heavily

affected the economic growth. Unexpected movements on oil price has effects on economic

stability for both supplier and producer countries, although more crucial for oil importing countries

with significant effects on the sales and profits of major industries worldwide. Therefore, modeling

and forecasting oil price are important to economic agents and policy makers especially given the

current COVID situation.

In reality, there are different types of crude oil – the thick, unprocessed liquid that drillers extract

below the earth – and some are more desirable than others. Thus, where the oil comes from also

makes a difference. Here, we use the most important two benchmarks, WTI and Brent, to

demonstrate the world crude oil price changes.

West Texas Intermediate (WTI) – WTI refers to oil extracted from wells in the U.S. and sent via

pipeline to Cushing, Oklahoma. The product itself is very light and very sweet, making it ideal for

gasoline refining, in particular. WTI continues to be the main benchmark for oil consumed in the

United States.

Brent Blend– Roughly two-thirds of all crude contracts around the world reference Brent Blend,

making it the most widely used marker of all. These days, “Brent” actually refers to oil from four

different fields in the North Sea: Brent, Forties, Oseberg and Ekofisk. Crude from this region is

light and sweet, making it ideal for the refining of diesel fuel, gasoline.
Time-Series Forecasting of WTI Crude and Brent Crude Oil Prices

In this project we have used two pronged approach a) Univariate b) Multivariate. In univariate we

try to predict the future Brent/WTI prices from their respective lagged values. In Multivariate

Analysis we try to predict the future oil prices and develop a regression model given both the WTI

and Brent oil prices. We follow a step by step approach by going from data pre-processing to

visualization, checking for stationarity, removing stationarity and appropriate tests to support

these. Further we go on to use the VAR model in Multivariate Analysis.

DATA AND METHODOLOGY

The data for this project is taken from the US Department of Energy, energy information

administration (EIA) independent statistics and analysis where it has a variety of information

available. The Spot Price of WTI and Brent Crude Oil is used to design the model. Monthly price

data from May, 1987 - Aug, 2020 comprising 200 observations is taken into account. The values

from May,1987- Oct,2019 are used to build the model and the last 10 values are used for validating

forecasting accuracy.

DATA VISUALIZATION

In order to get a basic knowledge of the crude oil price, we plot the daily WTI and Brent crude oil

price from 1987 to 2020. From the plot below, we find that WTI and Brent spot oil price almost

follows the same trajectory except for some small divergence from 2011 to 2014.

Summary Statistics of WTI and Brent Oil prices :-

WTI Crude Brent Crude

Summary Stats Oil ($) Oil($)

Min. value 11.35 9.82

1st Quartile 20.10 18.75

Time-Series Forecasting of WTI Crude and Brent Crude Oil Prices

Median 33.51 32.09

Mean 45.19 46.42

3rd Quartile 63.81 66.41

Max. value 133.88 132.72

Raw Plot Of WTI and Brent Prices

160
140
Prices in Dollars($)

120
100
80 WTI Prices
60 Brent prices
40
20
0
1988

1999

2001

2011
1987

1989
1990
1992
1993
1994
1995
1996
1997

2000

2002
2003
2004
2006
2007
2008
2009
2010

2013
2014
2015
2016
2017
2018
2020
The spot prices of crude oil have been profoundly influenced by events that have economic and

geo-political aspects as follows. These following insights can be seen from the graph plotted after

summary statistics of each of the Brent crude oil and WTI crude oil prices: -

❖ The remarkable price falls in the period 1997–1998, due to the slowdown of Asian

economic growth;

❖ OPEC (Organization of Petroleum Export Countries) curtailed the production of crude oil

by 4.2 million barrels per day between 2000 and 2001, resulting in an increase in crude

oil prices;

❖ In 2001-2003, 9/11 attacks and the invasion of Iraq raised concerns about the stability of

the Middle East’s production.

Time-Series Forecasting of WTI Crude and Brent Crude Oil Prices

❖ Then, Crude oil prices keep rising for a variety of reasons, including North Korea’s

missile launches, the crisis between Israel and Lebanon, and Iranian nuclear

brinkmanship.

❖ In 2008, The global financial crisis caused a bubble-bursting sell-off. Prices plummet

78.1% from July to December.

❖ In 2014, Strong production in the United States and Russia caused prices to crash from

July to December. OPEC’s November decision to maintain production further damages

the market heading into 2015.

❖ In 2015, U.S. output reached its highest level in more than 100 years. Prices hover near

$50 a barrel as of July 22.

UNIVARIATE ANALYSIS

To specify at the outset, we have used the Box-Jenkins method to do Univariate time series

analysis.

Box - Jenkins Analysis refers to a systematic method of identifying, fitting, checking, and using

integrated autoregressive, moving average

(ARIMA) time series models. This model

is a combination of the AR and MA models. It assumes that the time series is stationary. Box and

Jenkins recommend differencing non-stationary series one or more times to achieve stationarity.

Doing so produces an ARIMA model, with the "I" standing for "Integrated”. These models can be

extended to include seasonal autoregressive and seasonal moving average terms. The most general

Box-Jenkins model includes difference operators, autoregressive terms, moving average terms,

seasonal difference operators, seasonal autoregressive terms, and seasonal moving average terms
Time-Series Forecasting of WTI Crude and Brent Crude Oil Prices

The figures of the raw plot of WTI and Brent prices shows a non-stationary series. Clearly there is

a seemingly increasing trend coupled with fluctuations from 2002 after a leverage up and down

price observations from 1987 to 2002. The statistical properties like mean and variance are

changing over time.

BRENT CRUDE OIL

STAGE1- IDENTIFICATION

Following is the result of the Dicky Fuller Unit Root test on the raw data of Brent Crude Oil

AUGMENTED DICKEY-FULLER Test

Dickey-Fuller Lag- Order p-value

Brent Crude Oil -2.2416 7 0.4752
Raw Data

The p-value, 0.4752, clearly accepts the null hypothesis that the series is non-stationary. In order

to make it stationary, the first difference of the data is taken.

The figure shows the plot of first difference of Brent Crude oil datase

The graph of the 1 st differenced series looks smooth over the time period. It does not show any

presence of trend or fluctuating moments (especially mean) in the time series of order integrated=1.

Following is the result of the Dicky Fuller Unit Root test on the raw data.
Time-Series Forecasting of WTI Crude and Brent Crude Oil Prices

AUGMENTED DICKEY-FULLER Test

Dickey-Fuller Lag- Order p-value

Brent Crude Oil -7.9574 7 0.01
First Difference

The result of the Dicky Fuller test on the transformed data confirms that the new series is stationary.

From the p-value it is clear that the alternative hypothesis, that the series is

stationary, is accepted and the null hypothesis is rejected.

Now we go to the next steps within Stage 1 of identification which are finding the AR and MA

order: p and q respectively of our integrated time series. Once the p, q and d are found, we will

go to the next 2 steps which are estimation and diagnostic check.

For this we plotted the ACF and PACF of differenced time series of d=1
Time-Series Forecasting of WTI Crude and Brent Crude Oil Prices

Significant spikes are observed through lags 0,1,2,6,14 in the ACF pattern and through lags

0,6,13,25 in the PACF pattern give us the order of MA and AR respectively.

Time-Series Forecasting of WTI Crude and Brent Crude Oil Prices

STAGE 2 : ESTIMATION

1. AIC & BIC

AIC BIC
ARIMA(0,1,0) 2278.598 2282.562
ARIMA(0,1,1) 2233.243 2241.170
ARIMA(0,1,2) 2228.275 2240.165
ARIMA(0,1,6) 2231.047 2258.792
ARIMA(0,1,14) 2223.515 2282.969
ARIMA(6,1,0) 2229.854 2257.599
ARIMA(6,1,1) 2231.553 2263.261
ARIMA(6,1,2) 2232.984 2268.656
ARIMA(6,1,6) 2229.059 2280.585
ARIMA(6,1,14) - -
ARIMA(13,1,0) 2230.496 2285.986
ARIMA(13,1,1) 2231.335 2290.788
ARIMA(13,1,2) 2232.893 2296.311
ARIMA(13,1,6) 2220.441 2299.712
ARIMA(13,1,14) - -
ARIMA(25,1,0) 2235.845 2338.899
ARIMA(25,1,1) 2234.561 2341.578
ARIMA(25,1,2) 2229.997 2340.978
ARIMA(25,1,6) - -
ARIMA(25,1,14) - -

The Akaike information criterion (AIC) and Bayesian information criterion (BIC) are measures of

the relative goodness of fit of a statistical model similar to R square in cross sectional data series.

The preferred model is the one with the minimum AIC or BIC value. However, BIC value is

preferred over AIC value as BIC is a stronger tool which penalises for using higher lags again

similar to adjusted R square in multiple regression equation and data series.

We choose four models with minimum AIC and BIC i.e Model 3, Model 16, Model 17 and Model

18 and apply more tests and then select the best model. The model highlighted in red are not

considered as we are getting convergence error for the same and it’s beyond the scope of our study.
Time-Series Forecasting of WTI Crude and Brent Crude Oil Prices

STAGE 3: DIAGNOSTIC CHECK

1. RESIDUAL PLOT ANALYSIS:

Model 3- ARIMA(0,1,2)

Residuals of Model-3

We see that the ACF and PACF have some significant lags, therefore cannot conclude about the

white noise of the residuals of the model- ARIMA (0,1,2). Therefore, we plot the residual plot,

histogram of residuals and do an Augmented Dickey-Fuller test to check the white noise of the

model 3. We see that the histogram of residuals is centered around the mean 0 and is normally

distributed. The residual plot looks smooth and stationary.

Time-Series Forecasting of WTI Crude and Brent Crude Oil Prices

Following is the result of the Dicky Fuller Unit Root test on the residuals of Model-3.

AUGMENTED DICKEY-FULLER Test

Dickey-Fuller Lag- Order p-value

Residuals of -8.108 7 0.01
ARIMA(0,1,2)

The result of the Dicky Fuller test on the residual of Model 3-ARIMA (0,1,2) confirms that the

residuals of ARIMA (0,1,2) is stationary.

Model 16- ARIMA (25,1,0)

Residuals of Model-16

The ACF and PACF have no significant lags. But it is insufficient to conclude whether the residuals

of Model 16- ARIMA (25,1,0) is white noise. Therefore, we will plot the histogram of residuals,

time plot of residuals and check for stationarity using Augmented Dickey Fuller test.
Time-Series Forecasting of WTI Crude and Brent Crude Oil Prices

Following is the result of the Dicky Fuller Unit Root test on the residuals of Model-16.

AUGMENTED DICKEY-FULLER Test

Dickey-Fuller Lag- Order p-value

Residuals of -6.8372 7 0.01
ARIMA(25,1,0)

The result of the Dicky Fuller test on the residual of Model 3-ARIMA (25,1,0) confirms that the

residuals of ARIMA (25,1,0) is stationary.

MODEL-17 ARIMA (25,1,1)

Residuals of Model-17

The ACF and PACF have no significant lags. But as it is insufficient to conclude whether the

residuals of Model 17- ARIMA (25,1,1) is white noise. Therefore, we will plot the histogram of

residuals, time plot of residuals and check for stationarity using Augmented Dickey Fuller test.

Following is the result of the Dicky Fuller Unit Root test on the residuals of Model-17.
Time-Series Forecasting of WTI Crude and Brent Crude Oil Prices

AUGMENTED DICKEY-FULLER Test

Dickey-Fuller Lag- Order p-value

Residuals of -6.8579 7 0.01
ARIMA(25,1,1)

The result of the Dicky Fuller test on the residual of Model 3-ARIMA (25,1,0) confirms that the

residuals of ARIMA (25,1,1) is stationary.

MODEL-18 ARIMA (25,1,2)

Residuals of Model-18

The ACF and PACF have no significant lags. But as it is insufficient to conclude whether the

residuals of Model 18- ARIMA (25,1,2) is white noise. Therefore, we will plot the histogram of

residuals, time plot of residuals and check for stationarity using Augmented Dickey Fuller test.

Following is the result of the Dicky Fuller Unit Root test on the residuals of Model-18.
Time-Series Forecasting of WTI Crude and Brent Crude Oil Prices

AUGMENTED DICKEY-FULLER Test

Dickey-Fuller Lag- Order p-value

Residuals of -6.8796 7 0.01
ARIMA(25,1,2)

The result of the Dicky Fuller test on the residual of Model 3-ARIMA (25,1,2) confirms that the

residuals of ARIMA (25,1,2) is stationary.

By the residual plot analysis, we see that that all 4 models – ARIMA (0,1,2), ARIMA (25,1,0),

ARIMA (25,1,1), and ARIMA (25,1,2) are good for forecasting as all are white noise. We can now

forecast the values using all 4 models.

2. COMPARING FORECASTED AND ACTUAL VALUES

Actual ARIMA ARIMA ARIMA ARIMA

Values (0,1,2) (25,1,0) (25,1,1) (25,1,2)
Nov 2019 63.21 40.33045 37.07225 36.80330 37.54772
Dec 2019 67.31 39.72317 37.93861 37.88284 38.09959
Jan 2020 63.65 39.72317 40.37634 40.08855 40.21506
Feb 2020 55.66 39.72317 44.30426 44.05790 43.60920
Mar 2020 32.01 39.72317 44.95427 44.04298 43.35942
Apr 2020 18.38 39.72317 44.70931 44.13230 43.70720
May 2020 29.38 39.72317 44.43936 43.93708 43.82226
Jun 2020 40.27 39.72317 43.30310 42.74467 41.71262
Jul 2020 43.24 39.72317 43.90529 43.14156 42.50854
Aug 2020 44.74 39.72317 44.02978 43.43441 42.93894

The table provides the actual and forecasted values of the Brent Crude of months November 2019-

August 2020. To check the accuracy of the actual and predicted value, we get the root mean square

error for each of the model and select the model with lowest mean square error.

MODEL Actual
Values
ARIMA(0,1,2) 16.64579
ARIMA(25,1,0) 18.20493
Time-Series Forecasting of WTI Crude and Brent Crude Oil Prices

ARIMA(25,1,1) 18.11282
ARIMA(25,1,2) 17.86319

The RSME is lowest for the model ARIMA (0,1,2). The RMSE value of ARIMA (0,1,2) shows

that the forecast has been robust and indicating better fit of the model.

WTI CRUDE OIL

We follow the same process of identification of parameters of ARIMA (p, d, q), estimation of AIC

and BIC and selecting 4 models with lowest AIC and BIC, followed by diagnostic check of

residuals and forecasting of values. We further select the model - ARIMA (6,1,0) as it has the

lowest root mean square error for forecasting the values of WTI Crude Oil for the months of

Nov2019- Aug2020.
Time-Series Forecasting of WTI Crude and Brent Crude Oil Prices

The blue line above shows the forecasted for the month of Nov 2019 to Aug 2020. We observe that

there is some discrepancy in Oil prices between actual and forecasted values. This happens because

of the fact that the world is going through current pandemic situation coupled with recession. In

this part of univariate analysis, we consider only the lagged values of oil prices which is a limitation

in itself given the current economic shocks that the world is going through. Oil became the

buzzword overnight as benchmark US crude prices crashed below $0 for the first time in history

on 22April 2020, throwing investors around the world off guard. Shares of energy and oil firms

bled on stock markets around the world as experts predict a global energy glut in the wake of

suspended economy activity due to Covid-19 pandemic. Thus not only the lagged values but recent

demand shocks in crude oil because of slowing economic activity needs to be taken in to

consideration as the oil producing nations are running out of space to store oil, making oil

worthless in economy.
Time-Series Forecasting of WTI Crude and Brent Crude Oil Prices

MULTIVARIATE ANALYSIS

Vector auto regression (VAR) time series model is an econometric model used to capture the

evolution and the inter-dependencies between multiple time series. All the variables in a VAR are

treated symmetrically by including for each variable an equation explaining its evolution based on

its own lags and the lags of all the other variables in the model. Cointegration test and VECM

model could have been better improvement in our analysis as there were inherent drawbacks with

VAR but we have slightly tweaked our variables in Multivariate analysis to make the analysis

simple. We have considered forecasting ‘Growth rate of Crude-Oil prices’ rather than just

forecasting the prices as this transformed variable showed stationarity and was easy to interpret by

applying VAR models.

Our primary goal in the multivariate analysis are:

1. Estimating simultaneous equation model through Vector Auto-regressive (VAR) model

2. Checking response of one variable due to shock of another through Impulse Response

Function (IRF)

3. Verifying which variable’s lag causes others or not through Granger-Causality

4. Variance- Decomposition

5. Forecasting the variables ahead

We have seen in our univariate analysis that our data-sets of the price of both WTI and Brent are

non-stationary. And the basic assumption for VAR is that the repressors should be stationary. Hence

we will transform the initial variable of price to growth rate of price. Growth rate of price is given

by the formula,

𝑃𝑟𝑖𝑐𝑒 𝑖𝑛 𝑐𝑢𝑟𝑟𝑒𝑛𝑡 𝑝𝑒𝑟𝑖𝑜𝑑 − 𝑃𝑟𝑖𝑐𝑒 𝑖𝑛 𝑝𝑟𝑒𝑣𝑖𝑜𝑢𝑠 𝑝𝑒𝑟𝑖𝑜𝑑

𝑃𝑟𝑖𝑐𝑒 𝑖𝑛 𝑝𝑟𝑒𝑣𝑖𝑜𝑢𝑠 𝑝𝑒𝑟𝑖𝑜𝑑
Time-Series Forecasting of WTI Crude and Brent Crude Oil Prices

This transformation leads to finding the relationship between the price growth rate of both Brent

oil and WTI oil rather than the relationship between Brent price and WTI price.

STAGE 1-

Checking stationarity condition through Unit Root Test.

ADF Test for stationarity:

H0: Non-stationary

Ha: Stationary

Following is the result of the Dicky Fuller Unit Root test on the growth rate of Brent and WTI oil

price.

AUGMENTED DICKEY-FULLER Test

Dickey-Fuller Lag- Order p-value

Growth Rate of Brent Oil Price -8.0213 7 0.01

Growth Rate of WTI Oil Price -8.1318 7 0.01

Since the P-value of the Dickey-Fuller statistic is less than 0.05 & hence we can say that both

variables are stationary.

As a second step VAR Modelling is applied to the bi-variate series. One of the advantages

of VAR is that it addresses the problem of simultaneity in the bivariate model.

Time-Series Forecasting of WTI Crude and Brent Crude Oil Prices

● Plotting the bi-variate time series for VAR

Growth rates of Brent and WTI Crude Oil Price

0.6
0.4
0.2
g_brent_ts

0.0
-0.2
-0.4
0.6
0.4
g_WTI_ts

0.2
0.0
-0.2
-0.4

1990 1995 2000 2005 2010 2015 2020

Years

We can see that the raw plot of the bivariate is mean reverting which supports the claims of Unit

Root test for stationarity.

The next step is to find out the optimal lag length for the VAR model

AIC(n) HQ(n) SC(n) FPE(n)

4 2 1 4

The table shows that the optimal lag length for VAR is 4. Hence our model will be,

g_WTIt = WTI price growth rate

g_brentt = Brent price growth rate

g_WTIt = 𝛽 10+ 𝛽 11g_WTIt-1 +𝛽 12 g_WTIt-2+ 𝛽 13g_WTIt-3 +𝛽 14g_WTIt-4+𝛽 15g_brentt-1+

𝛽 16g_brentt-2+ 𝛽 17g_brentt-3+𝛽 18g_brentt-4 +ut -------------(i)

g_brentt = 𝛼10+𝛼 11g_WTIt-1 +𝛼12 g_WTIt-2+ 𝛼13g_WTIt-3+𝛼14g_WTIt-4+𝛼15g_brentt-1+

𝛼16g_brentt-2+ 𝛼17g_brentt-3+𝛼18g_brentt-4 +et -------------(ii)

Time-Series Forecasting of WTI Crude and Brent Crude Oil Prices

STAGE 2-VAR ESTIMATION

Endogenous variables: g_WTI, g_brent

Deterministic variables: const

Sample size: 395

Log Likelihood: 1204.526

Roots of the characteristic polynomial: 0.6713, 0.6713, 0.6331, 0,6331,0.5871, 0.5871, 0.5495,

0.5495

Estimation results for equation g_brent_ts:

g_brent_ts = g_brent_ts.l1 + g_WTI_ts.l1 + g_brent_ts.l2 + g_WTI_ts.l2 + g_brent_ts.l3 +

g_WTI_ts.l3 + g_brent_ts.l4 + g_WTI_ts.l4 + const

Estimate Standard Error t value Pr(>|t|)

g_brent_ts.l1 0.174620 0.158367 1.103 0.271
g_WTI_ts.l1 0.114189 0.164359 0.695 0.488
g_brent_ts.l2 -0.090090 0.157780 -0.571 0.568
g_WTI_ts.l2 -0.056285 0.163312 -0.345 0.731
g_brent_ts.l3 0.041708 0.157616 0.265 0.791
g_WTI_ts.l3 -0.043030 0.162460 -0.265 0.791
g_brent_ts.l4 0.017973 0.160654 0.112 0.911
g_WTI_ts.l4 -0.171294 0.170940 -1.002 0.317
const 0.006796 0.004866 1.397 0.163

This results shows there are no statistically significant coefficients which mean that there will be

no significant effect on the WTI price growth rate by the Brent price growth rate in the long-run.
Time-Series Forecasting of WTI Crude and Brent Crude Oil Prices

Estimation results for equation g_WTI_ts:

g_WTI_ts = g_brent_ts.l1 + g_WTI_ts.l1 + g_brent_ts.l2 + g_WTI_ts.l2 + g_brent_ts.l3 +

g_WTI_ts.l3 + g_brent_ts.l4 + g_WTI_ts.l4 + const

Estimate Std. Error t value Pr(>|t|)

g_brent_ts.l1 0.235151 0.152869 1.538 0.125
g_WTI_ts.l1 0.026887 0.158652 0.169 0.866
g_brent_ts.l2 0.121853 0.152301 0.800 0.424
g_WTI_ts.l2 -0.228009 0.157642 -1.446 0.149
g_brent_ts.l3 0.108581 0.152143 0.714 0.476
g_WTI_ts.l3 -0.150723 0.156820 -0.961 0.337
g_brent_ts.l4 0.137960 0.155076 0.890 0.374
g_WTI_ts.l4 -0.268752 0.165005 -1.629 0.104
const 0.005835 0.004697 1.242 0.215

Again the result shows there are no statistically significant coefficients which mean that there will

be no significant effect on the brent price growth rate by the WTI price growth rate in the long-

run.

From this we cannot conclude about the long-run relationship between WTI and Brent price, which

can be shown through the better model VECM.

But the VAR model is stable as we can see that all the roots of the characteristic polynomial are

less than one (within the unit root circle).

STAGE 3- DIAGNOSTICS TESTS

Serial Correlation between the error terms

Portmanteau Test (asymptotic)

Data: Residuals of VAR Object Model

Chi-squared = 46.369 df = 48 p-value = 0.5399

The above table shows that there is serial correlation between the error terms which means there

is autocorrelation between the error terms. Hence evidence in favor for white noise errors in this

stationary bi-variate series can be provided at a very high level of significance.

Time-Series Forecasting of WTI Crude and Brent Crude Oil Prices

Stability of VAR

From the above graph we can see that the fluctuation is within the band (in red color). Hence our

VAR model is stable and we can proceed further with the VAR(4) model.

Granger Causality Block Significance Test

Model 1: g_WTI_ts ~ Lags(g_WTI_ts, 1:4) + Lags(g_brent_ts, 1:4)

Model 2: g_WTI_ts ~ Lags(g_WTI_ts, 1:4)

Res.Df Df F Pr(>F)
1 386
2 390 -4 0.8673 0.4836

The Granger Causality Test of VAR tells about the correlation between the two variables in this

bi-variate series. The F-test does not give a strong correlation between WTI price growth rate and

Brent price growth rate.

Time-Series Forecasting of WTI Crude and Brent Crude Oil Prices

Impulse Response Function

This graph helps to understand how a variable responds or behaves in the subsequent period to the

sudden change of another variable. The shocks can be in the form of some out of control random

event.

● Shock in Brent price growth rate

We can see that for the first 3 periods after the shock there will be decline in WTI price growth

rate but will be positive which imply that there will be rise in WTI price but at declining rate then

for next 5 periods the growth rate will be negative and eventually converge to 0 which mean in

long run no price change in the long-run.

Time-Series Forecasting of WTI Crude and Brent Crude Oil Prices

● Shock in WTI price growth rate

From the above graph, we can conclude that the shock will lead to cycle type fluctuation in the

Brent price growth rate.

Variance Decomposition

Here, the variance decomposition tells how the variance in explanatory variables affects the

variance or the variation in dependent variables and for how long in the future.
Time-Series Forecasting of WTI Crude and Brent Crude Oil Prices

So, variance in Brent growth rate explains the variance in WTI price growth rate at a very high

degree of about 0.9 or 90% and for 9 future periods; this is persistent/prolonged impact. However,

this is not the case where variance in WTI price growth rate explains variance in Brent price growth

rate as we can see in the first diagram.

STAGE 4-FORECASTING AND PLOTING OF FORECAST

VAR has two main purposes: forecasting and creating relationships of the simultaneous

equations, especially when we don’t have our data in differenced form but rather in levels, as the

case here with both variables integrated at order I(0).

The forecasts and the relationship established by means of Granger Causality, Impulse Response

Function and Variance Decomposition is meticulously done throughout the project with one

assumption of everything else being the same.

With this, we end VAR bi-variate analysis.

Time-Series Forecasting of WTI Crude and Brent Crude Oil Prices

WAY FORWARD

Although this model has helped examining the relationship between variables and its forecasting,

it fails when there is a non-stationary series. Forecasting would be difficult in such a case as data

will be in differences then. And this is the limitation of our multivariate analysis as initially our

data sets are non-stationary. Hence this analysis can be better modelled when we include the

cointegration test and VECM model. As mentioned before, in our univariate analysis the other

factors that can be taken into consideration given the COVID-19 and recessionary situations are

demand shocks in the world economy, OPEC oil prices etc. Thus bringing these factors in to our

model would make our model more comprehensive and accurate in forecasting prices. This is an

aspect which can be looked forward to as future extension of our current analysis.