Technology forecasting using DEA in the presence of infeasibility

• Published date: 01 September 2018
• Author: Dong‐Joon Lim
• DOI: http://doi.org/10.1111/itor.12198
• Date: 01 September 2018

Intl. Trans. in Op. Res. 25 (2018) 1695–1706

DOI: 10.1111/itor.12198

INTERNATIONAL

TRANSACTIONS

IN OPERATIONAL

RESEARCH

Technology forecasting using DEA in the presence of

infeasibility

Dong-Joon Lim

Department of Engineering and Technology Management, PortlandState University, Portland OR, USA

E-mail: dongjoon@pdx.edu [Lim]

Received 12 June2014; received in revised form 14 June 2015; accepted 27 June 2015

Abstract

As a predictive applicationof data envelopment analysis (DEA), technology forecastingusing DEA (TFDEA)

measures the rate of frontier shift by whichthe arrival of future technologies can be estimated. However, it is

well known that DEA and therefore TFDEA may sufferfrom the issue of infeasible super-efficiency especially

under the condition of variable returns to scale. This study develops an extended TFDEA model based on

the modified super-efficiency model proposed in the literature, which has the benefit of yielding radial super-

efficiency scores equivalent to those obtained from the original super-efficiency model when feasibility is

present. The previously published applicationof liquid crystal displays (LCD) is revisited to illustrate the use

of the new model. The results show that the proposed approach makes a reasonable forecast for formerly

infeasible targets as well as a consistent forecast for feasible targets.

Keywords:technological forecasting; data envelopment analysis; infeasibility; rate of change

1. Introduction

Data envelopment analysis(DEA) studies, originated by Charnes et al. (1978), have often examined

the changing performance of decision making units (DMUs) over time (Charnes et al., 1984;

Sueyoshi, 1992; Fare et al., 1994; Cooper et al., 2006; Emrouznejad and Thanassoulis, 2010). To

put it plainly, let θt

0(xt

0,yt

0)denote the efficiency of the DMUoin time period twith input and

output characteristics (xt

0,yt

0)being measured against the frontier of peers also in time period

t. The efficiency of that DMU compared against peers from the following period would then be

θt+1

0(xt

0,yt

0). Under the input-oriented (IO) model, if the optimal value of θt+1

0(xt

0,yt

0)is less than

1.0, the unit in period tis inefficient relative to units from period t+1. In contrast, if the optimal

value of θt+1

0(xt

0,yt

0)is equal to 1.0, then the unit in period toutperforms units from period t+1in

some manner and is efficient.

C

2015 The Authors.

International Transactionsin Operational Research C

2015 International Federation of OperationalResearch Societies

Published by John Wiley & Sons Ltd, 9600 Garsington Road, Oxford OX4 2DQ, UK and 350 Main St, Malden, MA02148,

USA.