Restructuring performance prediction with a rebalanced and clustered support vector machine

• Date: 01 July 2018
• Author: Lu‐Yao Hong,Yu‐Chang Mo,Pei‐Chann Chang,Bang‐Zhu Zhu,Hui Li
• DOI: http://doi.org/10.1002/for.2512
• Published date: 01 July 2018

RESEARCH ARTICLE

Restructuring performance prediction with a rebalanced

and clustered support vector machine

Hui Li

1

|Lu‐Yao Hong

2

|Yu‐Chang Mo

3

| Bang‐Zhu Zhu

4

| Pei‐Chann Chang

5

1

College of Tourism and Service

Management, Nankai University, Tianjin

300350, China

2

Department of Business, Tobacco

Monopoly Administration of Tiantai,

Taizhou, Zhejiang 317200, China

3

School of Mathematical Sciences,

Huaqiao University, Quanzhou, Fujian

362021, China

4

Management School, JiNan University,

Guangzhou, Guangdong 510632, China

5

Department of Information

Management, Yuan Ze University, Chunli

32026, Taiwan

Correspondence

Hui Li, College of Tourism and Service

Management, Nankai University, Tianjin

300350, China.

Email: lihuihit@live.cn

Funding information

National Natural Science Foundation of

China, Grant/Award Number: 71571167

Abstract

This paper discusses whether asset restructuring can improve firm performance

over decades. Variation in the stock price or the financial ratio is used as the

dependent variable of either short‐or long‐term effectiveness to evaluate the

variance both before and after asset restructuring. The result is varied. It is nec-

essary to develop a foresight approach for the mixed situation. This work pio-

neers to forecast effectiveness of asset restructuring with a rebalanced and

clustered support vector machine (RCS). The profitability variation 1 year

before and after asset restructuring is used as the dependent variable. The cur-

rent financial indicators of the year of asset restructuring are used as indepen-

dent variables. Specially treated listed companies are used as research samples,

as they frequently adopt asset restructuring. In modeling, the skew distribution

of samples achieving and failing to achieve performance improvement with

asset restructuring is handled with rebalancing. The similar experienced knowl-

edge of asset restructuring to the current asset restructuring is filtered out with

clustering. With the help from rebalancing and clustering, a support vector

machine is constructed for prediction, together with other forecasting models

of multivariate discriminant analysis, logistic regression, probit regression,

and case‐based reasoning. These models'standalone modes are used as

benchmarks. The empirical results demonstrate the applicability of the RCS

for forecasting effectiveness of asset restructuring.

KEYWORDS

effectivenessof asset restructuring, merger and acquisition, prediction, rebalancing and clustering

classification

1|INTRODUCTION

Asset restructuring is assumed to be one of the most effec-

tive methods for improving a firm's performance and

helping the firm recover from distressed conditions. In

China, asset restructuring commonly includes equity

transfer, asset stripping, asset acquisition, asset replace-

ment, merger, and debt restructuring. With these opera-

tions, a restructured firm is expected to significantly

modify its debt, operations, and structure to eliminate

financial harm and improve business performance. After

an asset restructuring event, the departments, ownership,

operations, working processes, or main business may

change. As a result, the restructured firm will become

more integrated and profitable, and have a better chance

to avoid bankruptcy. There are several advantages that

make asset restructuring frequently adopted by firms.

Asset restructuring is an effective means to quickly

achieve economies of scale, with which the cost will be

reduced. Competition among the same industry can be

Received: 17 February 2015 Revised: 14 September 2017 Accepted: 2 December 2017

DOI: 10.1002/for.2512

Journal of Forecasting. 2018;37:437–456. Copyright © 2018 John Wiley & Sons, Ltd.wileyonlinelibrary.com/journal/for 437

reduced with asset restructuring, as a result of which the

organizational efficiency will be improved. By merging

corporates with diverse business, the operation risk will

be reduced. Thus asset restructuring is frequently

employed by firms with the expectation to achieve perfor-

mance improvement.

For a specific firm, however, there is no absolute assur-

ance that asset restructuring will result in improved per-

formance. Here, two methods are created to identify

whether asset restructuring is an effective method for firms

to improve their performance. The first method is called

event approach and is based on marketing information,

which uses variations in stock price before and after asset

restructuring as the dependent variable to evaluate effec-

tiveness of asset restructuring. This method attempts to

determine whether an abnormal return is achieved follow-

ing an asset‐restructuring announcement. However, it is

difficult to distinguish whether the abnormal return ana-

lyzed using an event approach is the result of value

enhancement caused by asset restructuring or by market

mispricing (Healy, Palepu, & Ruback, 1992). Thus, in place

of an abnormal return, post‐restructuring accounting and

cash‐flow return are used to directly test the change in

long‐term operational performance. The second method

is called accounting approach. It provides performance

indicators created from key financial ratios. The second

method determines whether financial performance indica-

tors improve after asset restructuring. Following a discus-

sion of effectiveness of asset restructuring based on

testing the two methods (i.e., event approach and account-

ing‐based performance evaluation approach) using vari-

ous samples (e.g., from the USA or the UK) over decades

(Cheng & Leung, 2004), we can find that there is no

consensus on whether asset restructuring confers benefits

that lead to either abnormal returns or financial returns.

To help managers make efficient decisions related to

asset restructuring in the current situation of our mixed

understanding of effectiveness of asset restructuring, this

work adopts a third method of restructuring performance

prediction (RPP) of corporate assets. To help firms predict

the possibility of performance improvement after asset

restructuring, we aim to develop a forecasting method

that considers not only the balance between samples that

achieve and samples that fail to achieve performance

improvement with asset restructuring but also the simi-

larity between the current problem and past cases. The

possibility of predicting improved performance after asset

restructuring enables a judgement of whether companies

will improve their performance within the expected

timeframe after adopting specific asset‐restructuring strat-

egies. It is especially urgent for distressed firms to predict

the performance of asset restructuring because such firms

frequently employ asset restructuring.

The possibility of predicting improved performance

after asset restructuring is a new topic in the literature

and refers to predicting the possibility of improved perfor-

mance after adopting specific asset‐restructuring strate-

gies. The statistical model of multivariate discriminant

analysis (MDA) is useful for this task (Li & Zhang,

2012). MDA is a very widely used model that finds a linear

combination of variables that separates two or more clas-

ses of objects. Its advantages include its ease of under-

standing, interpretation, and application, along with its

acceptable level of performance. Even in a situation in

which the assumed normal distributions of variables do

not occur, MDA is capable of generating high perfor-

mance. MDA remains a leading model in practical

applications of bankruptcy prediction (Altman, 1968),

face recognition (McLachlan, 2004), and earth science

(Perriere & Thioulouse, 2003), among others. MDA is a

changeable model for forecasting effectiveness of asset

restructuring. Samples from firms that have engaged in

asset restructuring commonly violate MDA's normality

assumptions, resulting in a weak theoretical foundation

for forecasting effectiveness of asset restructuring. Thus

models that require fewer assumptions than MDA—e.g.,

logistic regression (logit) and probit regression, among

others—are useful in RPP to solidify our theoretical foun-

dation. In addition to the model requirements issue, two

other issues are not considered in developing an effective

forecasting approach for asset restructuring in emerging

markets such as China. One issue involves the relatively

small volume and skew of samples, which could poten-

tially decrease the reliability and generalizability of

findings. The second issue involves the small volume of

available candidate variables, which potentially misses

some significant variables for this task. The task of asset

restructuring involves both financial and operational

aspects. To solidify the findings for effectively predicting

asset restructuring, more useful variables, samples, and

models with fewer assumptions should be considered.

All three of the above‐cited issues—model requirements,

samples, and variables—relate to the modeling process.

There is no consideration of how the current problem

behaves. The model is separated from the characteristics

of the target problem that the model is supposed to solve.

Thus this is the fourth issue: The link between a model

and its target problem should be considered to generate

a more precise prediction of effectiveness of asset

restructuring.

To consider all of the four issues—i.e., the model

requirements issue, the sample issue, the variable issue,

and the linkage issue—we attempt to propose a new

method to forecast effectiveness of asset restructuring by

integrating rebalancing, clustering, and a support vector

machine (SVM). A larger sample asset‐restructuring set

438 LI ET AL.