An Application of Combination Prediction Model and Macroscopic Analysis in Predicting Asset-liability Ratio

An Application of Combination Prediction Model and
Macroscopic Analysis in Predicting Asset-liability Ratio

Yue He, Dan Zhang, Yujie Cao
(Business School of Sichuan University, Chengdu 610064, China)

hechangzheng@scu.edu.cn

Abstract: This paper analyzed the significance of predicting the asset-liability ratio and the importance of
choosing macroscopic data to formulate a model. It used GMDH predicting model. AC predicting model. ARCH model
with macroscopic data and combination predicting model to predict the asset-liability ratio of Sichuan industrial
enterprises. It analyzed and authenticated the results of the four predicting models. At last. it evaluated the effect of the
four models in predicting the asset-liability ratio.

Key Words:
GMDH Model; AC Model; ARCH Model; Combination Predicting Model; Asset-liability Ratio
1 Introduction
The asset-liability ratio reflects the size of assets proportion offered by creditors in the whole assets of enterprises.
and it also reflects the risk procedures of offering enterprises credit funds by creditors. and it reflects the management
and business ability of enterprises by borrowing debt[1]. If the production and operational capability of enterprises is
good and the operating condition is at a good state. the asset-liability ratio will be at a higher level. On the contrary. the
asset-liability ratio will be relatively low. The prediction of asset-liability ratio is an important means for enterprises to
predict its development potential. and it is also one of the important methods for government to master regional
macroeconomic situation. In so many prediction methods. GMDH predicting model and AC predicting model in Self-
organizing Data Mining theory and Method have been a hotspot in macroeconomic prediction research. National
Bureau of Statistics formally introduced prosperity survey methods to national statistical report systems. So prosperity
data is quite important for predicting macroeconomic indicators[2][3].
This paper used GMDH predicting model. AC predicting model. ARCH model with prosperity data and
combination predicting model to predict the asset-liability ratio of Sichuan industrial enterprises. It analyzed and
authenticated the results of the four predicting models in order to use more appropriate predicting model to predict
macroeconomic indicators.

2 Predicting Model and Empirical Analysis

5.1 GMDH auto-regressive prediction model

The earliest auto-regressive data mining-Group Method of Data Handling was first put forward by
A.G.Ivakhnenko academician from Ukraine Academy of Sciences. GMDH auto-regressive prediction model was
produced by the combination of GMDH algorithm thought and auto-regressive thought. GMDH predicting method only
uses predicted indicator data which doesn’t include other indicator data. It takes the historical data of predicting
indicator as input variables. and it screen models by using external criterion until obtaining an optimal model.
There are seven implementation steps in GMDH auto-regressive algorithm[4][5][6].
Let Sichuan enterprises asset-liability ratio data from the first season of 2002 to the third season of 2006 to be the
training set of this model. which includes 19 data points. And the fourth season of 2006 to the fourth season of 2007 to
be the detection set of predicting results. Use GMDH model to predict and use Knowledge Miner software to deal with
y
y
y
data and get predicting model 1 : 1 = 1.000e+ 1 (t-1) + 9.999e+1. The predicting results are shown in Table 1.
2.2 AC model algorithm
AC (Analog Complexing) algorithm is Analog Complexing merging algorithm. It was put forward and developed
by Lorence. AC algorithm can be taken as a sequence model recognition method of predicting. clustering and
classifying complex objects. Generally speaking. AC algorithm includes three steps:

289

---

1) the production of patterns to be selected: To a given m-dimensional sequence with N observed value
x =
t
{x , ,x
1t L mt }(t =1,2, N)
L
P (i)
. the definition of a pattern is from the i row with k-row table k
. and this k is
(i =1,2, ,N ? k +1)
called the length of the pattern
L
.
2) the change of patterns to be selected;
x
= ai + ai x
, j = 0,1, ,k ?1;i =1,2, ,N ? k +1;l =1,2, , .
m
i
+
i
l, j
+
i
l, j
L
L
L
a
Let
0l
1l
. parameter 0l can be
i
P (i)
a
explained to be the state difference between referenced pattern and similar pattern k
. and parameter 1l can be
explained to be some uncertain factors.
3) the choose of patterns to be selected: The main purpose of this step is recognizing the similarity between
P (i)
patterns. We call the similarity as pattern similarity. In order to measure the similarity of a to-be-selected pattern k

R
on referenced pattern P
which has been changed as step (2). it need to measure the distance between k observed
values with m system variables in the two patterns.
Let Sichuan enterprises asset-liability ratio data from the first season of 2002 to the third season of 2006 to be the
training set of this model. which includes 19 data points. And the fourth season of 2006 to the fourth season of 2007 to
be the detection set of predicting results. Use AC model to predict and use Knowledge Miner software to deal with the
data and the results of this predicting model are shown in Table 1.
2.3 ARCH model theory and the steps of modeling
ARCH model is called autoregressive conditional heteroscedasticity. It was put forward by Engle in 1982.
Autoregressive Conditional Heteroskedasticity model is very important in the study of nonlinear time series. It can take
all the available information as conditions. and use some auto-control way to reflect the difference of variance. ARCH
model is usually used in modeling random disturbance term of agent model. It can extract the information of residual
effectively. and it offered a method which uses past error to explain the future predicting error. This paper analyzed the
correlation between prosperity data and asset-liability ratio data by analyzing the prosperity index and asset-liability
ratio data, and then used prosperity data as independent variables to determine their correlation coefficients. And then it
establish model to predict asset-liability ratio.
The several steps of the process of ARCH predicting model based on prosperity data in Sichuan enterprises asset-
liability ratio predicting are as follows:
(1) the correlation analysis between asset-liability ratio and enterprises prosperity data;
(2) using obviously correlative variables from step (1) to establish corresponding ARCH model;
(3) the detection and analysis of established model;
(4) using the established model to predict asset-liability ratio and make some analysis.
This paper chose the prosperity index indicators which have high correlation with asset-liability ratio from the 1st
season of 2002 to the 4th season of 2007 to establish ARCH model. In order to eliminate the influence of some factors
such as price index, this paper computes the original data to ratio data, and uses the ratio data to make correlation
analysis. The prosperity data this paper chosen were as follows: entrepreneur confidence index, enter price prosperity
index, the rate of increase of GDP index, production prosperity index, capital prosperity status, enterprises financing
prosperity status.
According to the four steps of ARCH analysis, using SPSS statistical software to establish ARCH model on asset-
liability ratio [8]. and the results of ARCH model are shown in Table 1.
2.4 Combination predicting model
Combination predicting is a method that combinates several different predicting methods properly by using the
information offered by these methods comprehensively in order to improve the prediction accuracy. Professor
C.Granger, who is the 2003 Nobel Laureate in economics and from university of California in USA, evaluated on
combination predicting model that combination predicting model offers a simple and practical method which may
produce better forecast results.
Suppose that there are n predicting models established to predict the asset-liability ratio (in this paper n=3). The
f ( )
t , f ( )
t
f ( )
t
forecast value on goal variable of each model is respectively 1
2
L n , and then the combination predicting

290

---

n
f (t) = ?? f (t) + c
i i
? ? ?
?
model is
i 1
=
. Where c is a constant, 1, 2, 3, L n are the weights of the predicting value of
each prediction method in combination predicting model. The determination of constant c and the weights
? (i=1,2,...,n )
i
is according to the principle of least square method.
Let the prediction results of GMDH auto-regressive prediction model, AC model, ARCH model apply into the
combination predicting model, and then use SPSS software to solve this problem to get the value of constant c and the
? (i=1,2,...,n )
weights i
. The values are as follow: .
c = 80.475553,? = 0.209209,
?
? = ?0.128178,? = 0.046507
1
2
3

Put the value of parameters into the combination predicting model (h) and get the prediction results in Table 1.

2.5 Predicting results

Table 1: The results of the four prediction models
GMDH Predicting
AC Predicting
ARCH Predicting
combination predicting
Actual

value (%)
Predicting
Predicting
Predicting
Predicting Predicting Predicting Predicting Predicting
value
error
value
error
value
error
value
error
the 4th
season of
62.26 62.92 0.66 62.44 0.17 64.29
2.02 62.29
0.03
2006
the 1st
season of
62.06 63.02 0.96 63.09
1.03
61.39 -0.66
62.05 -0.01
2007
the 2nd
season of
62.58 62.50 -0.08 62.78 0.19 62.94
0.35 62.27
-0.31
2007
the 3rd
season of
62.21 62.47 0.26 62.39 0.18 62.28
0.06 62.30
0.09
2007
the 4th
season of
62.07 62.44 0.36 63.04 0.97 62.99
0.92 62.25
0.18
2007
Data resource: Statistical Bureau of Sichuan Province

3 The results analysis of each prediction model

Each of the four prediction models has its own character in predicting the asset-liability ratio of Sichuan
enterprises. And the forecast results of each prediction model are in Table 2.
Table 2: the prediction effect comparison of several prediction models
the 4th
the 1st
the 2nd
the 3rd
the 4th
absolute
relative

season
season of
season of
season of
season of
average
average
of 2006
2007
2007
2007
2007
error
error
GMDH
prediction
0.7524%
0.6614 0.9653 -0.0859 0.2627 0.3663 0.4683
error
AC prediction
error
0.1773 1.0336 0.1981 0.1830 0.9725 0.5129 0.8241%
ARCH
prediction
2.0288 -0.6687 0.3572 0.0673 0.9186 0.8081 1.2985%
error
Combination
prediction
0.0349 -0.0027 -0.3097 0.0905 0.1826 0.1241 0.1993%
error

291

---

From Table 2, in the prediction of the asset-liability ratio of Sichuan enterprises, the combination model got the
best prediction effect. And the absolute average error between its forecast value and its actual value is only 0.1241, and
the prediction error of each year keeps at a 0-0.3 fluctuation level, and the relative average error is only 0.1993%, which
conveyed that combination prediction model had high prediction accuracy in predicting the asset-liability ratio of
Sichuan enterprises.
The effect of GMDH prediction model is worse than combination prediction. The absolute average error is 0.4683,
and the relative average error is only 0.7524%. So GMDH is one of the proper methods to predict the asset-liability
ratio of Sichuan enterprises.
AC prediction model and GMDH prediction model are both self-organization data mining models. Their similarity
in predicting the same group of data was proofed in this paper. From Table 2, the absolute average error of AC
prediction model is 0.5129, and the relative average error is 0.8241%, the relative average error is fluctuated in 1%
level. So it conveyed that the two self-organization data mining models both had effective results in predicting the asset-
liability ratio of Sichuan enterprises.
From Table 2, the effect of ARCH prediction model based on prosperity data in this paper is not as good as other
methods. This had a different conclusion with other scholars on GDP[7], the industrial value added etc. prediction using
ARCH prediction model based on prosperity data. The reason may be that the macroeconomic indicators such as GDP,
the industrial value added etc. are impacted by macroeconomic policies and macroeconomic development trend greatly.
The prosperity index is gained by surveying enterprisers on the macroeconomic policies and macroeconomic
development trend. But the asset-liability ratio is a relative indicator. Its impaction of macroeconomic policies and
macroeconomic development trend is not great as absolute indicators.
To sum up, in predicting macroeconomic indicators we should choose proper prediction models according to the
significance of indicators to get satisfactory prediction effect. And then the prediction can help government and related
departments to make policies and adjust policies.

4. Conclusions
This paper used GMDH predicting model, AC predicting model. ARCH model with prosperity data and
combination predicting model based on least square method and chose the data from the 1st season of 2002 to the 4th
season of 2007 to predict the asset-liability ratio of Sichuan industrial enterprises and authenticate the results.
The results of empirical research showed that:
1) Combination prediction model had high prediction accuracy and good prediction effect on predicting the asset-
liability ratio of Sichuan enterprises, which is the same with other scholars’ research on combination prediction in
predicting economic indicators such as GDP[7] etc. This conveyed that combination prediction model is a relatively
better model in economy prediction.
2) We should choose different prediction models according to different economic indicators. The use of ARCH
prediction model with prosperity data in this paper showed that the effect of its application in predicting the asset-
liability ratio of Sichuan industrial enterprises is not ideal, but its prediction effect on GDP[7] is quite good. So it’s
necessary to choose different prediction models according to different economic indicators.
The much better economic prediction model in this paper has been used in the word to predict the asset-liability
ratio of Sichuan industrial enterprises and has got good effects on prediction.

References
[1] Zhong Chen. Huimin Ji.Analysis Enterprises Asset-liability Ratio[J].China Economist.2001(3):108-109
[2] Lei Chen. Enterprises Situation and Macro-economy [J]. Management World. 2004(3):14-24
[3] Yue He. etc. the Application of Macro-economy Prosperity Prediction Model Based on GMDH [J]. Statistics and
Decision. 2007(6):52-54
[4] Changzhen He. Auto Organization Data Mining and Economy Prediction [M]. Beijing: Science Press. 2005.9
[5] Mueller J.A.. Lemke Frank. Self-organizing data mining [M]. Hamburg. Libri. 2000
[6] Ivakhnenko A. G..Kozubovski S. F.Yu. V. Kostenko. Objective System Analysis of Macroeconomic Systems [J].
Syst. Anal. Model. Simul. I. 1990. 3:201-201
[7] Shiqiang Wei. etc. GDP Combination Prediction based on ARMA-ARCH [J]. Statistics and Decision. 2007(5):21-
22
[8] Qinguo Ma. Management and Statistics-Data Getting. Statistics Theory. SPSS tool and Application Research [M].
Beijing: Science Press. 2002.

292

---