Market Intelligence and Price Forecast

34: Market Intelligence And Price Forecast
MARKET INTELLIGENCE AND PRICE FORECAST

S.P. BHARDWAJ
IASRI, New Delhi 110012
Market Intelligence (MI) is knowledge based management system which may be defined as a
process primarily based on market information collected over period of time. An analysis based
on past information helps to take decision about the future. MI synthesizes information from
many diverse sources to form greater insights. MI requires sophisticated understanding of
strategic trade goals and to widen trade opportunities. The MI provides diversified avenues to
examine the market beaviour of agricultural commodities to facilitate all the stake holders.
Market integration and Price Forecast, the two important components of Market Intelligence are
chosen for discussion in this study The issue of market integration has become an important area
of empirical research in development economics. Many developing economies have been
implementing structural adjustment and market reform programmes. An important component of
such programme is the liberalization of commodity markets. It has been argued that such
liberalization is required for achieving allocative efficiency and long-term growth in agriculture.
It has also been argued that freeing domestic and external trade in food grains from government
intervention, while maintaining its role in price stabilization, yields positive welfare benefits.
However, unless food markets are spatially integrated, producers and consumers will not realize
the gains from liberalization: the correct price signals will not be transmitted through the
marketing channels, farmers will not be able to specialize according to long-term comparative
advantage and the gains from trade will not be realized. Spatial market integration refers to a
situation in which the prices of a commodity in spatially separated markets move together and
price signals and information are transmitted smoothly. Spatial market performance may be
evaluated in terms of a relationship between the prices of spatially separated markets, and spatial
price behavior in regional markets may be used as a measure of overall market performance.
Price is the primary mechanism by which various levels of the market are linked. The extent of
adjustment and speed with which shocks are transmitted among producer, wholesale, and retail
market prices is an important factor reflecting the actions of market participants at different
levels. Price forecasting has been very important in decision making at all levels and sectors of
the economy. In agriculture, where the decision environment is characterized by risks and
uncertainty largely due to uncertain yields and relatively low price elasticities of demand of the
most commodities, decision makers require some information about the future and the likelihood
of the possible future outcomes. Price forecasts are critical to market participant making
production and marketing decisions and to policy makers who administer commodity programs
and assess the market impacts of domestic or international events. Therefore commodity price
movements have a major impact on overall macroeconomic performance. Hence, commodityprice forecasts are a key input to macroeconomic policy planning and formulation.
Review of Literature
Erik Hjalmarsson and Par Osterholm (2007) have investigated the properties of Johansens
(1988, 1991) maximum eigenvalue and trace tests for cointegration under the empirically
1
relevant situation of near-integrated variables. Using Monte Carlo techniques, we show that in a
system with near-integrated variables, the probability of reaching an erroneous conclusion
regarding the cointegrating rank of the system is generally substantially higher than the nominal
size. The risk of concluding that completely unrelated series are cointegrated is therefore nonnegligible. The spurious rejection rate can be reduced by performing additional tests of
restrictions on the cointegrating vector(s), although it is still substantially larger than the nominal
size.
Megha Mukim, Karan Singh, A Kanakaraj (2009) they have examined whether the wheat
market is integrated across states in India, and concludes that the market is integrated in the long
run. This long run integration, however, does not come from the free flow of goods across states
in the country, but from the sharing of similar production technologies by farmers across states.
The paper also shows that the market for wheat is not integrated in the short run. This implies
that at a given time period there exist two prices for the same commodity, since transaction costs
are the main barriers to market integration. The paper also estimates such transaction costs using
transport and communication infrastructure indices across states, and concludes that there exist
large variations resulting in high transaction costs.
Madhusudan Ghosh (2003) has investigated intra-state and inter-state spatial integration of
wheat markets in India. In view of the limitations of the methods used earlier for investigating
market integration in Indian agriculture, this study has utilized the ML method of cointegration.
Intra-state regional integration of wheat markets has been evaluated by testing the linear long-run
relationship between the prices of the state-specific variety of wheat quoted in spatially separated
locations in five selected states. The cointegration results for Bihar and UP indicate that the
regional wheat markets were integrated to such an extent that the weak version of the LOP was
in operation. The cointegration tests also offer evidence for regional wheat market integration in
Haryana, Punjab and Rajasthan; but no evidence is found in favour of the LOP for these states.
The results for inter-state regional wheat markets represented by five market centres chosen from
the five selected states reveal three cointegrating vectors and two common stochastic trends.
Contrary to Jha et al. (1997), we find that though all the prices taken together are integrated, they
are not pair-wise cointegrated.
Christopher B. Barrett (2005) considered that markets aggregate demand and supply across
actors distributed in space. Well-integrated markets play a fundamental role in ensuring that
macro level economic policies change the incentives and constraints faced by micro-level
decision-makers, in distributing risk and in preserving incentives to adopt improved production
technologies. Yet the literature is replete with evidence of forgone arbitrage opportunities in both
intra- and inter-national trade. Given limited data and the restrictive assumptions of existing
empirical methods, economists still have only a fragile empirical foundation for reaching clear
judgements about spatial market integration as a guide for corporate or government policy. The
literature on price forecasting has focused on two main classes of linear, single-equation,
reduced-form econometric models as well as Time Series models. The first group (Financial
Models) includes models which are directly inspired by financial economic theory and based on
the market efficiency hypothesis (MHE), while models belonging to the second class (Structural
Models) consider the effects of commodity market agents and real variables on commodity
prices.
2
Reza Moghaddasiand and Bita Rahimi Badr(2008) considered wheat (Bread) is a dominant
product in the consumption basket of Iranian households and can be considered as a strategic
commodity. In this paper different econometric models including structural and time series
models are specified and estimated. The literature on price forecasting has focused on two main
classes of linear, single-equation, reduced-form econometric models as well as Time Series
models. The first group (Financial Models) includes models which are directly inspired by
financial economic theory and based on the market efficiency hypothesis (MHE), while models
belonging to the second class (Structural Models) consider the effects of commodity market
agents and real variables on commodity prices. Then forecasting performance of these models
are evaluated and compared by using common criteria such as: root mean square error, mean
absolute error, mean absolute percentage error and Theil inequality coefficient. The data used in
this study include annual farm and guaranteed prices of wheat and rice(as a competitive product)
and wheat stock for 1966 to 2006.Main findings reveal the superiority of time series models(unit
root and ARIMA(3,2,5)) for forecasting of wheat price. ARIMA) annual models outperformed
the structural model in predicting the price of wheat for the period 1966-2006. The unit root and
ARIMA models were also constructed using only the information provided by the historical time
series of the variable being forecast; hence the amount of information required to develop these
models was considered to be less than those employed in formulating the structural models.
Likewise the costs involved in developing the econometric structure forecasting models were
considered to be more than the cost associated in developing time series models. It is difficult to
conclude about the adequacy of the forecasts derived from the selected models, since it largely
depends on the particular use to which the price predictions are to be employed.
Rangsan Nochai and Titida Nochai (2006) This research is a study model of forecasting oil
palm price of Thailand in three types as farm price, wholesale price and pure oil price for the
period of five years, 2000 2004. The objective of the research is to find an appropriate ARIMA
Model for forecasting in three types of oil palm price by considering the minimum of mean
absolute percentage error (MAPE). The results of forecasting were as follows: ARIMA Model
for forecasting farm price of oil palm is ARIMA (2,1,0), ARIMA Model for forecasting
wholesale price of oil palm is ARIMA (1,0,1) or ARMA(1,1), and ARIMA Model for
forecasting pure oil price of oil palm is ARIMA (3,0,0) or AR(3) . In this paper, we developed
model for three types of oil palm price, were found to be ARIMA(2,1,0) for the farm price
model, ARIMA(1,0,1) for whole sale price, and ARIMA(3,0,0) for pure oil price. Which we can
see that the MAPE for each model very small.
Chakriya Bowman and Aasim M. Husain (2004) the paper aims to assess the accuracy of
alternative price forecasts for 15 primary Commodities over the past decade. A number of
alternate measures of forecast performance, having to do with statistical as well as directional
accuracy, are employed. The analysis indicates that although judgmental forecasts tend to
outperform the model-based forecasts over short horizons of one quarter for several
commodities, models incorporating futures prices generally yield superior forecasts over
horizons of one year or longer. Spot and futures prices were generally found to be nonstationary
and, in most cases, spot and futures prices appear to be cointegrated. Although there is
considerable comovement between spot and futures prices, futures prices tend to exhibit less
variability than spot prices. Hence, futures prices tend to act as an anchor for spot prices, and
error- correction models that exploit the long-run cointegrating relationship provide better
3
forecasts of future spot-price developments. When evaluating the ex-post effectiveness of

forecasts, standard statistical measures are commonly used. Mean pricing error, mean absolute
pricing error, mean absolute relative pricing error (MARPE), median absolute relative pricing
error and root mean squared error (RMSE) are typically calculated and the results used to
generate conclusions about the accuracy of forecasts. This research will focus primarily on
RMSE, which gives a measure of the magnitude of the average forecast error, as an effectiveness
measure. The ECM forecasts outperform the other types of forecasts for eight of the fifteen
commodities at the eight quarter horizon. In some of these cases, the ECM forecast performance
is superior in both statistical and directional terms (wheat, soybeans, and soybean meal),
although for several commodities the ECM yields significantly better directional accuracy at the
expense of somewhat lower statistical accuracy (aluminum, lead, nickel, zinc, and maize). For
another four commodities (tin, soybean oil, sugar, and cotton), the ECM performs about as well
as judgment at the eight-quarter horizon, and both perform better than the best unit root/ARMA
forecasts.
K. Assis, A. Amran, Y. Remali and H. Affendy, (2010) the purpose of this study was to compare
the forecasting performances of different time series methods for forecasting cocoa bean prices.
The monthly average data of Tawau cocoa bean prices graded SMC 1B for the period of January
1992-December 2006 was used. Four different types of univariate time series methods or models
were compared, namely the exponential smoothing, autoregressive integrated moving average
(ARIMA), generalized autoregressive conditional heteroskedasticity (GARCH) and the mixed
ARIMA/GARCH models. Root mean squared error (RMSE), mean absolute percentage error
(MAPE), mean absolute error (MAE) and Theil's inequality coefficient (U-STATISTICS) were
used as the selection criteria to determine the best forecasting model. This study revealed that the
time series data were influenced by a positive linear trend factor while a regression test result
showed the non-existence of seasonal factors. Moreover, the Autocorrelation function (ACF) and
the Augmented Dickey-Fuller (ADF) tests have shown that the time series data was not
stationary but became stationary after the first order of the differentiating process was carried
out. Based on the results of the ex-post forecasting (starting from January until December 2006),
the mixed ARIMA/GARCH model outperformed the exponential smoothing, ARIMA and
GARCH models.
Padhan, P.C., (2012) considered that forecasting of any issues, events or variables requires an indepth understanding of the underlying factors affecting it. Such is the case for forecasting annual
productivity of agricultural crops. Agricultural productivity, in the context of India, extensively
depends upon numerous factors namely: good rainfall, timely use of appropriate fertilizer and
pesticides, favorable climate and environments, agricultural subsidies given to farmers etc.
Therefore, forecasting productivity of agricultural crops is not only tedious but also
indispensible, as large chunk of people depends on agriculture for their livelihood. Various univariate and multi-variate time series techniques can be applied for forecasting such variables. In
this paper, ARIMA model has been applied to forecast annual productivity of selected
agricultural product. For empirical analysis a set of 34 different products has been considered,
contingent upon availability of required data. Applying annual data from 1950 to 2010,
forecasted values has been obtained for another 5 years since 2011. The validity of the model is
verified with various model selection criteria such as Adj R2, minimum of AIC and lowest
MAPE values. Among the selected crops, tea provides the lowest MAPE values, whereas
cardamom provides lowest AIC values.
Liew Khim Sen, Mahendran Shitan and Huzaimi Hussain ( 2007) It is important to forecast
price, as this could help the policy makers in coming up with production and marketing plan to
improve the Sarawaks economy as well as the farmers welfare. In this paper, we take up time
series modelling and forecasting of the Sarawak black pepper price. Our empirical results show
that Autoregressive Moving Average (ARMA) time series models fit the price series well and
they have correctly predicted the future trend of the price series within the sample period of
study. Amongst a group of 25 fitted models, ARMA (1, 0) model is selected based on postsample forecast criteria.
Guillermo Benavides (2009) the author suggested that there has been substantial research effort
aimed to forecast futures price return volatilities of financial and commodity assets. Some part of
this research focuses on the performance of time-series models (in particular ARCH models)
versus option implied volatility models. A significant part of the literature related to this topic
shows that volatility forecast accuracy is not easy to estimate regardless of the forecasting model
applied. This paper examines the volatility accuracy of volatility forecast models for the case of
corn and wheat futures price returns. The models applied here are a univariate GARCH, a
multivariate ARCH (the BEKK model), an option implied and a composite forecast model. The
composite model includes time-series (historical) and option implied volatility forecasts. The
results show that the option implied model is superior to the historical models in terms of
accuracy and that the composite forecast model was the most accurate one (compared to the
alternative models) having the lowest mean-square-errors. Given these findings it is
recommended to use a composite forecast model if both types of data are available i.e. the timeseries (historical) and the option implied. In addition, the results of this paper are consistent to
that part of the literature that emphasizes the difficulty on being accurate about forecasting asset
price return volatility. This is because the explanatory power (coefficient of determination)
calculated in the forecast regressions were relatively low.
Kailash Chandra Pradhan and K. Sham Bhat (2009) The study investigated price discovery,
information and forecasting in Nifty futures markets. Johansens (1988) Vector Error Correction
Model (VECM) is employed to investigate the causal relationship between spot and futures
prices. This study compares the forecasting ability of futures prices on spot prices with three
major forecasting techniques namely ARIMA, VAR and VECM model. The results indicate that
spot market leads the futures market and spot market serves as a primary market for price
discovery. The leading role of futures market weakens around the firm specific announcements
(Mukherjee and Mishra, 2006). Futures market is now in immature stage, which has been started
from June 2000. Still many traders and investors are confused about this new market. Derivatives
are complex. The payoffs and risk that buyer and seller face are considerably more difficult than
those seen on the equity market. The new traders and investors are still facing difficulty to entry
in the futures market. Therefore, spot market leads futures market. Also, the findings suggest that
vector error correction model (VECM) performs well on a post-sample basis against the
univariate auto regressive integrated moving average (ARIMA) model and a vector auto
regression (VAR) model. The results show clearly that it is important to take into account the
long-run relationship between the futures and the spot prices in forecasting future spot prices.
HYPOTHESIS
The study based on following hypothesis
1. The time series price data used in the study has unit roots or non-stationary
2. The spatial markets are cointegrated in the long run
3. Price forecast using GARCH and ARIMA methods
DATA & METHODOLOGY
This study focuses on the Box-Jenkins and GARCH methods of forecast Gram price in spot
markets. The study has been illustrated with the time series data on Spot price of Gram in Delhi,
Indore and Bikaner Markets from 01 January 2007 to 19 April 2012.
Statioinarity Test:
The stationary of the data series is evaluated by Augmented Dickey-Fuller (ADF) tests. The most
widely used tests for unit roots are Dickey and Fuller (1979) test and the Augmented Dickey
Fuller (ADF) test. Both are used to test the null hypothesis that the series has unit root or
non-stationary.
The DF Test is stated as follows:
Y t Y t 1 e t
(1)
Where and are parameters and et is random term.

Here the null hypothesis is that H0 : = 1 indicating that the series is non-stationary.
Y
t1
..(2)
Where = - 1 & Yt = Yt - Yt-1

The null hypothesis is H0 : = 0.
The test can be carried out by performing a -test on the estimated . The - statistics under the
null hypothesis of a unit root does not follow the conventional t- distribution. Dickey and Fuller
(1979) showed that distribution under null hypothesis is non-standard and simulated critical
values for selected samples size. If the error term et is auto-correlated, the equation (2) is
modified as
t 1
m
Yt 1
t
i
i1
(3)
Where m = number of lagged difference terms required so that the error term t is serially
independent. The null hypothesis is the same as the DF test, i.e., H0 : = 0, implying that Yt is
non-stationary. When DF test is applied to models like the equation (3), it is called Augmented
Dickey Fuller (ADF) test. The augmented DickeyFuller (ADF) statistic, used in the test, is a
negative number. The more negative it is, the stronger the rejection of the hypothesis that there is
a unit roots at some level of confidence.
Johansen's cointegration tests: Cointegration of prices between two markets can be studied
after establishing the order of integration for the price series. The prices are subjected pair wise
to linear regression.
Yt
(4)
Where Yt & Xt = price series & ut = error term.

The residuals (ut) of the equation (4) are subjected to DF or ADF test. The test statistic is
compared to the critical values provided by Engle and Granger (1987). If the residuals are
stationary, then the series are co integrated or otherwise. Engle and Granger (1987) pointed out
that a linear combination of two or more non-stationary series may be stationary. If such a
stationary linear combination exists, the non-stationary time series are said to be cointegrated.
The stationary linear combination is called the cointegrating equation and may be interpreted as a
long-run equilibrium relationship among the variables. The purpose of the cointegration test is to
determine whether a group of non-stationary series is cointegrated or not. As explained below,
the presence of a cointegrating relation forms the basis of the Vector Error Correction (VEC)
specification. Vector Auto Regressive (VAR) based cointegration tests using the methodology
developed by Johansen (1991, 1995) has been used in this study.
Consider a VAR of order :

------------------------------(5)
Where
is a k-vector of non-stationary I(1) variables,
is a d-vector of deterministic
is a vector of innovations. We may rewrite this VAR as,
variables, and
- ------------------- ------- (6)

Where:
------------(7)
Granger's representation theorem asserts that if the coefficient matrix has reduced rank r < k ,
and
is I(0).
then there exist k r matrices and each with rank r such that
r is the number of cointegrating relations (the cointegrating rank) and each column of is the
7
cointegrating vector. The elements of are known as the adjustment parameters in the VEC
model. Johansen's method is to estimate the
matrix from an unrestricted VAR and to test
whether we can reject the restrictions implied by the reduced rank of .
NUMBER OF COINTEGRATING RELATIONS
The first test statistic (trace) tests whether the number of distinct cointegrating vectors is less
than or equal to r. The second test statistic (max) tests the null that the number of cointegrating
vectors is r against an r+1. Johansen and Jueselius (1990) provided the critical values of these
statistics. The trace statistic tests the null hypothesis of r cointegrating relations against the
alternative of k cointegrating relations, where k is the number of endogenous variables, for
. The alternative of
cointegrating relations corresponds to the case
where none of the series has a unit root and a stationary VAR may be specified in terms of the
levels of all of the series. The trace statistic for the null hypothesis of r cointegrating relations
is computed as:
---------- (8)
Where
is the i-th largest eigenvalue of the
matrix. The maximum eigenvalue statistics
which tests the null hypothesis of

cointegrating relations against the alternative of
cointegrating relations. This test statistic is computed as:
----------(9)
for
Forecast Methods
Time Series Models
The price forecasts based on these models are non-structural-mechanical forecasts and economic
theory and related issues are not being followed. The primary concept in the analysis of time
series involves the basic tools of probability theory. The simplest form of a forecasting model is
the unit root model with drift, which may be written asPt = + P t-1 + et
Where Pt is the commodity price at time t and P t-1 pertains to lagged price. The error term, et, is
assumed to be white noise. If the price series contain a unit root, then a difference stationary
model should be used to model prices, otherwise the basic stationary model is appropriate.
Autoregressive integrated moving average (ARIMA) models are a class of linear models that are
capable of representing stationary as well as non-stationary time series. Since Gram prices are
volatile over the time trend, a heteroscedasticity approach shall be tested for the entire data
series. Hence, we use a GARCH model which is able to capture volatility in time series prices.
Its performance is then compared with ARIMA model.
The approach to forecasting is based on Box and Jenkins (1970) popularly known as ARIMA
model. The ARIMA is an extrapolation method, which requires historical time series data of
8
underlying variable. The methodology refers to the set of procedures for identifying, fitting, and
checking ARIMA models with time series data.
Non-seasonal Box-Jenkins Models for Stationary Series

(1) A pth order autoregressive model: AR(p), which has the general form:
Yt
1 Y t 1 2 Y t 2 ..........
.....
Yt
------(10)
Yt Response (dependent) variable at time t

Yt 1 , Yt 2, ........., Yt p = Response variable at time lags t-1, t-2, ., t-p, respectively.
0 ,1 , 2 ,......... ., p Coefficients to estimated, t Error term at time t.

(2) A qth-order moving average model: MA(q), which has the general form:
Y t t 1 t 1 2 t 2 .......... ....... q t q
--------------(11)
Where, Yt Response (dependent) variable at time t
Constant mean of the process, 1 , 2 ,..........., q Coefficients to be estimated
t Error term at time t., t 1 , t 2 ,......... .., t q Errors in previous time periods that are
incorporated in the response Yt.
(3) Autoregressive Moving Average Model: ARMA(p,q), which has general form:
Yt 0 1Yt 1 2Yt 2 .................... pYt p t 1 t 1 2 t 2 ........... q t q

ARIMA model-building
Model Identification - Determine whether the series is stationary or not by considering the
graph of ACF. If a graph of ACF of the time series values either cuts off fairly quickly or
dies down fairly quickly, then the time series values should be considered stationary.
According to equation (2.1), a highly useful operator in time-series theory is the lag or
backward linear operator (B) defined by
BYt = Yt-1
Model for non-seasonal series are called Autoregressive integrated moving average model,
denoted by ARIMA ( p, d, q). Here p indicates the order of the autoregressive part, d
indicates the amount of differencing, and q indicates the order of the moving average part. If
the original series is stationary, d = 0 and the ARIMA models reduce to the ARMA models.
The difference linear operator (), defined byYt Yt Yt 1 Yt BYt (1 B )Yt
The stationary series Wt obtained as the dth difference ( d ) of Yt

W
d Y t (1 B ) d Y t
ARIMA (p,d,q) has the general form:
( B )( 1 B ) d Y t
or
( B )
( B )W t q ( B ) t
Model Checking
In this step, model must be checked for adequacy by considering the properties of the residuals
whether the residuals from an ARIMA model must has the normal distribution and should be
random. An overall check of model adequacy is provided by the Ljung-Box Q statistic. The test
statistic Q is
m
r 2 (e)
Q m n (n 2) k
~ X m2 r
n
k
k 1
Where rk(e) = the residual autocorrelation at lag k
n= the number of residuals
m= the number of time lags includes in the test.
If the p-value associated with the Q statistic is small (p-value < ), the model is considered
inadequate. The analyst should consider a new or modified model and continue the analysis until
a satisfactory model has been determined.
GARCH Method
In econometrics, AutoRegressive Conditional Heteroskedasticity (ARCH) models are used to
characterize and model observed time series. They are used whenever there is reason to believe
that, at any point in a series, the terms will have a characteristic size, or variance. In particular
ARCH models assume the variance of the current error term to be a function of the actual sizes
of the previous time periods' error terms: often the variance is related to the squares of the
previous innovations.
Such models are often called ARCH models (Engle, 1982), although a variety of other acronyms
are applied to particular structures of model which have a similar basis. ARCH models are
employed commonly in modeling financial time series that exhibit time-varying
volatility clustering, i.e. periods of swings followed by periods of relative calm. If
an autoregressive moving average model (ARMA model) is assumed for the error variance, the
10
model is a Generalized
Bollerslev(1986)) model.
Autoregressive
Conditional
Heteroskedasticity (GARCH,
The GARCH (1, 1) Model Specification:

To measure the extent of price volatility GARCH (1, 1) Model has been applied in the study
-------------------------(12)
---------------(13)
The mean equation given in equation (10) is written as a function of exogenous variables with an
error term. Since
is the one-period ahead forecast variance based on past information, it is
called the conditional variance. The conditional variance equation specified in equation (13) is a
function of three terms:
A constant term:
.
News about volatility from the previous period, measured as the lag of the
squared residual from the mean equation:
(the ARCH term).
Last period's forecast variance:
(the GARCH term).
The (1, 1) in GARCH (1, 1) refers to the presence of a first-order autoregressive GARCH term
(the first term in parentheses) and a first-order moving average ARCH term (the second term in
parentheses). An ordinary ARCH model is a special case of a GARCH specification in which
there are no lagged forecast variances in the conditional variance equation-i.e., a GARCH (0, 1).
This specification is often interpreted in a financial context, where an agent or trader predicts this
period's variance by forming a weighted average of a long term average (the constant), the
forecasted variance from last period (the GARCH term), and information about volatility
observed in the previous period (the ARCH term). If the asset return was unexpectedly large in
either the upward or the downward direction, then the trader will increase the estimate of the
variance for the next period. This model is also consistent with the volatility clustering often seen
in financial returns data, where large changes in returns are likely to be followed by further large
changes. There are two equivalent representations of the variance equation that may aid you in
interpreting the model:
(1) If we recursively substitute for the lagged variance on the right hand side of equation(13)
we can express the conditional variance as a weighted average of all the lagged squared
residuals:
-------------- (14)
We can see that the GARCH (1, 1) variance specification is analogous to the sample
variance, but it down-weights more distant lagged squared errors.
11
(2) The error in the squared returns is given by

. Substituting for the variance
in the variance equation and rearranging terms we can write our model in terms of the
errors:
----------------------(15)
Thus, the squared error follow a hetroskedastic ARMA (1,1) process. The autoregressive root
which governs the persistence of volatility shocks is the sum of plus . The ARCH parameters
corresponds to and GARCH parameters to . If the sum of ARCH and GARCH coefficients
close to 1, indicating that volatility shocks are quite persistent.
Absolute Accuracy Performance Measures of Forecast
The absolute accuracy analysis is the statistic, root mean squared error (MSE), defined as: MSE
= (t yt)2 , Where yt and t are the actual and forecast values, respectively. MSE is considered
as a "non-parametric" statistic that indicates the size of the individual forecast errors from actual
values. The square root of MSE, called the root mean squared error (RMSE) represents the mean
size of forecast error, measured in the same units as the actual values
T h
RMSE =
( y
t T 1
yt ) 2
The absolute size of the errors the mean absolute forecast error (MAE) is used:
T h
MAE =
t T 1
yt / h
The RMSE is similar to MAE. The MAE and RMSE depend on the scale of the dependent
variable. These should be used as relative measures to compare forecasts for the same series
across different models.
The MAPE is calculated using the following formula
T h
MAPE = 100
t T 1
yt yt
/h
yt
The MAPE calculates the forecast error as a percentage of actual value. The drawback of the
MAPE is that it puts a heavier penalty on the forecasts that exceed the actual value and on those
that fall behind the actual value. The MAPE is similar to MAE except that it is dimensionless. It
will be helpful in making comparison among forecasts from different situations. For instance, to
compare forecasting methods in two different situations with different units of measure, one can
calculate the MAPEs and then average across situation. When the cost of errors is more closely
related to the percentage error than to the unit error, the MAPE is appropriate.
Empirical results:
12
Unit Root Test

Augmented Dickey Fuller (ADF) test was applied to the Spot price series data to test the null
hypothesis that the series has unit root or non- stationary. The results are given in Table-1. The
-Statistics obtained for all the price series is significant and greater than at 1 percent level,
the null hypothesis of series has unit root or non- stationary data series cannot be rejected. The
alternative hypothesis is true. Thus data series is subjected to first differencing to make the data
stationary. The results of differenced series indicated that the -Statistic obtained for price
series is not significant and less than at 1 percent level, we are bound to reject the null
hypothesis and the alternative hypothesis of stationary series and no unit root is true. The data
series became stationary at one differencing and the data is now ready for further econometric
analysis.
Johansens Cointegration Test
The cointegration in Spot market price of Delhi, Indore and Bikaner was carried out for Gram
(Chana) and the results are given in Table-2. In case of trace statistics cointegration r denotes the
number of cointegrating vectors. In this case, the number of cointegrating vectors can be at most
three as there are four series in each contract. Trace statistics show that the null hypothesis of
spot prices are not cointegrated (r=1) against the alternative of more cointegrating vectors (r>1)
is rejected. Next, the null hypothesis of r = 1 against the alternative of more cointegrating vectors
cannot be rejected at 5 percent significance level for all the cases. The presence of cointegrating
vectors in all cases shows that there exists long run relationship between spot prices. Similarly
Eigen value statistics rejects the null hypothesis of equal to or less than one cointegrating
equation against the alternative hypothesis of more than one equation. The two statistical tests
confirm the relationship of short and long run in the spot market price in all the three markets.
The same has been depicted with the help of cointegration graph.
Table-1. Augmented Dickey Fuller Test for Spot Market Price.
ADF Test value

1% level
5% level
10% level
ADF Test value
1% level
5% level
10% level
ADF Test value
Delhi Market
Level Data
At First Difference
t-Statistic
Prob.* t-Statistic
Prob.*
-1.787078
0.7108
-25.52475
0
-3.963866
-3.963866
-3.412658
-3.412658
-3.128297
-3.128297
Indore Market
-1.168269
0.9155
-25.91525
0
-3.963866
-3.963866
-3.412658
-3.412658
-3.128297
-3.128297
Bikaner Market
-1.658755
0.769
-38.71645
0
13
1% level
5% level
10% level
-3.963855
-3.412653
-3.128293
14
-3.963859
-3.412654
-3.128294
Table-2. Market Cointegration test for Spot market price, Delhi, Indore and Bikaner
Series Spot Price Gram in :Delhi Indore and Bikaner markets
Lags interval (in first differences):
1 to 4
Unrestricted CointegrationRank Test Trace Statistics
Hypothesized
Trace
0.05
Critical
No. of CE(s)
Eigenvalue Statistic Value
Prob.**
None *
0.267073 1181.358 42.91525
1
At most 1 *
0.229323
691.37 25.87211
0
At most 2 *
0.162993 280.5834 12.51798
0
Unrestricted Cointegration Rank Test Eigenvalue Statistics
MaxHypothesized
Eigen
0.05
Critical
No. of CE(s)
Eigenvalue Statistic Value
Prob.**
None *
0.267073 489.9877 25.82321 0.0001
At most 1 *
0.229323 410.7866 19.38704 0.0001
At most 2 *
0.162993 280.5834 12.51798
0
Cointegration Graph
300
300
200
200
100
100
-100
-100
-200
-200
-300
-300
250
500
750
1000
DST_DELHI
DST_INDORE
DSTBIKNER
15
1250
1500
Table-3. GARCH results for price forecast.

Method: ML - ARCH (Marquardt) - Normal distribution
Included observations: 1582 after adjustments
Convergence achieved after 32 iterations
MA Backcast: 1
GARCH = C(3) + C(4)*RESID(-1)^2 + C(5)*GARCH(-1)
Variable
Coefficient
Std. Error
z-Statistic
Prob.
C
0.510687
0.895857
0.570054
0.5686
MA(1)
0.136026
0.027407
4.963155
0
Variance Equation
C
RESID(-1)^2
GARCH(-1)
R-squared
Adjusted R-squared
S.E. of regression
Sum squared resid
Log likelihood
F-statistic
Prob(F-statistic)
Inverted MA Roots
17.46498
0.095291
0.895813
0.011176
0.01055
37.17141
2183108
-7730.87
4.464328
0.00137
-0.14
3.640192
0.008898
0.007871
Mean dependent var
S.D. dependent var
Akaike info criterion
Schwarz criterion
Hannan-Quinn criter.
Durbin-Watson stat
Actual, forecasted and residual values graph GARCH

300
200
100
0
300
-100
200
-200
100
-300
0
-100
-200
250
500
Residual
750
1000
Actual
16
1250
Fitted
1500
4.79782
10.70901
113.807
0
0
0
0.632111
37.36906
9.779861
9.796822
9.786163
2.051967
Table-5. ARIMA results for forecast

Dependent Variable: Spot Market Price Delhi
Method: Least Squares
Sample (adjusted): 5 1583
Included observations: 1579 after adjustments
Convergence achieved after 6 iterations
MA Backcast: 4
Variable
Coefficient Std. Error
t-Statistic Prob.
C
0.715621
0.933213
0.766836
0.4433
AR(3)
-0.10086
0.02506 -4.024721
0.0001
MA(1)
0.10425
0.025065
4.159212
0
R-squared
0.021553
Mean dependent var
0.715643
Adjusted R-squared
0.020312
S.D. dependent var
37.3522
S.E. of regression
36.97091
Akaike info criterion
10.06004
Sum squared resid
2154153
Schwarz criterion
10.07023
Log likelihood
-7939.4
Hannan-Quinn criter.
10.06383
F-statistic
17.35824
Durbin-Watson stat
2.008001
Prob(F-statistic)
0
Inverted AR Roots
.23+.40i
.23-.40i
-0.47
Inverted MA Roots
-0.1
Inverse Roots of ARIMA
17
Inverse Roots of AR/MA Polynomial(s)

-1.5
1.5
AR roots
MA roots
-1.0
-0.5
0.0
0.5
1.0
1.5
1.5
1.0
1.0
0.5
0.5
0.0
0.0
-0.5
-0.5
-1.0
-1.0
-1.5
-1.5
-1.5
-1.0
-0.5
0.0
0.5
1.0
1.5
Actual, forecasted and residual values graph ARIMA

300
200
100
0
300
-100
200
-200
100
-300
0
-100
-200
250
500
750
Residual
1000
Actual
1250
1500
Fitted
Table 5 : Information criterion for ARIMA and GARCH models

Model
ARIMA
GARCH
AIC
10.06004
9.779861
SIC
10.07023
9.796822
18
The AIC and SIC values are obtained from equation estimation from both ARIMA and GARCH
models using EViews. We found that both the AIC and SIC values from GARCH model are
smaller than that from ARIMA model. Therefore, it shows that GARCH is a better model than
ARIMA for estimating daily prices.
Table-6. Forecast Performance of ARIMA method
Forecast
Days
5
10
15
20
30
45
60
MAE
MAPE
RMSE
ARIMA GARCH ARIMA GARCH ARIMA GARCH
18.9335 14.8622
0.5583
0.4454 29.8229 19.4706
43.3339 41.9636
1.2197
1.1894 73.1369 71.9955
44.3099 40.0949
1.2372
1.1321 67.7201 63.2775
44.2905 44.5485
1.2163
1.2347 63.7383 63.0461
47.3229 46.0093
1.3023
1.2690 62.9799 60.7677
51.5128 50.2203
1.4255
1.3869 65.9775 64.3158
47.5725 48.2899
1.3248
1.3451 61.5202 61.8682
All forecast errors from GARCH model are smaller than that from ARIMA model. Therefore,
we can conclude that GARCH model performs better than ARIMA.
References
Johansen, S. (1988), Statistical Analysis of Cointegrating Vectors, Journal of Economics
Dynamic and control, 12, pp. 231-254.
Johansen, S. and K. Juselius (1990), Maximum Likelihood Estimation and Inference on
cointegration with application to the Demand for Money, Journal of Econometrics,
53, pp.211-244.
Johansen, Soren (1991). "Estimation and Hypothesis Testing of Cointegration Vectors in
Gaussian Vector Autoregressive Models," Econometrica, 59, 1551-1580
Johansen, Soren (1995). Likelihood-based Inference in Cointegrated Vector Autoregressive
Models, Oxford: Oxford University Press.
Erik Hjalmarsson and Par Osterholm (2007) Testing for Cointegration Using the Johansen
Methodology when Variables are Near-Integrated. International Finance Discussion
Papers Number 915, Board of Governors of the Federal Reserve System.
Megha Mukim, Karan Singh, A Kanakaraj (2009). Market Integration, Transaction Costs
and the Indian Wheat Market: A Systematic Study. Economic & Political Weekly
EPW may 30, vol xliv no 22.
Kailash Chandra Pradhan and K. Sham Bhat (2009) An Empirical Analysis of Price
Discovery, Causality and Forecasting in the Nifty Futures Markets. International
Research Journal of Finance and Economics. Euro Journals Publishing, Inc. ISSN
1450-2887 Issue 26 pp83-92.
Madhusudan Ghosh(2003) Spatial Integration of Wheat Markets in India: Evidence from
Cointegration Tests, Oxford Development Studies, Vol. 31, No. 2,
Christopher B. Barrett (2005) Spatial Market Integration. The New Palgrave Dictionary of
Economics, 2nd Edition (London: Palgrave Macmillan, orthcoming).
19
Reza Moghaddasiand and Bita Rahimi Badr(2008) An Econometric Model for Wheat Price
Forecasting in Iran .Paper presented in International Conference on Applied
Economics ICOAE.
Chakriya Bowman and Aasim M. Husain(2004) Forecasting Commodity Prices: Futures
Versus Judgment. IMF Working Paper. WP/04/41.March.
K. Assis, A. Amran, Y. Remali and H. Affendy, (2010) A Comparison of Univariate Time
Series Methods for Forecasting Cocoa Bean Prices. Trends in Agricultural
Economics, 3: 207-215.
Liew Khim Sen, Mahendran Shitan and Huzaimi Hussain ( 2007) Time Series Modelling
and Forecasting of Sarawak Black Pepper Price. MPRA Paper No. 791, posted 07.
November / 01:16.
Padhan, P.C., (2012) Application of ARIMA model for forecasting agricultural productivity
in India. J. Agric. Soc. Sci., 8: 5056.
Rangsan Nochai and Titida Nochai (2006) Arima Model for Forecasting Oil Palm price.
Paper presented in IMT- GT regional conference on Mathematics.Statistics and
Applications. Universiti Sains Malasyia, Penang June-13-15.
Box, G. and G. Jenkins, 1970. Time Series Analysis: Forecasting and Control. San
Francisco: Holden-Day.
20

Market Intelligence and Price Forecast

Enviado por

Dados do documento

Direitos autorais

Formatos disponíveis

Compartilhar este documento

Compartilhar ou incorporar documento

Opções de compartilhamento

Você considera este documento útil?

Este conteúdo é inapropriado?

Direitos autorais:

Formatos disponíveis

Market Intelligence and Price Forecast

Enviado por

Direitos autorais:

Formatos disponíveis

34: Market Intelligence And Price Forecast