6. Total Factor Productivity and Resource Reallocation 135 1. Measurement of Total Factor Productivity Growth
6.2.1. Previous studies on total factor productivity growth
TFP growth in Singapore attracted attention after two studies by Young (1992) and Kim and Lau (1994) that found “zero” TFP growth. In
addition, based on the empirical findings of these two studies, Krugman (1994) criticized the rapid economic growth performance of Singapore by claiming that it was based on factor accumulation like in the former Soviet Union rather than TFP gains and further claimed that such high growth rates cannot be sustained. The Asian financial crisis in 1997 in Southeast Asia increased the interest in these studies.
There are various studies on TFP growth in Singapore. However, the findings of these studies are controversial; for the same period some studies find positive TFP growth rate while others find a negative one. The main reason for these differences is different methodologies employed. Fur- thermore, capital stock values are not available from official statistics and have to be estimated by researchers. Differing estimates of capital stock result in different capital and TFP growth rates. Previous studies on TFP growth in Singapore used one of the three approaches: (i) conventional growth accounting approach that computes TFP growth rate as Solow residual, (ii) econometric methods such as the stochastic frontier approach, and (iii) translog production function. A list of some previous TFP growth estimates in the past is presented in Table 6.2.
Those studies using the conventional growth accounting approach compute TFP growth as a residual by deducting weighted growth rates of capital and labor input from growth rate of real value-added, the weights being the respective shares of capital and labor in value-added. Their findings are rather optimistic compared to the results of other studies, pro- ducing generally positive TFP growth rates. In addition, various studies point to improvements in TFP performance for the aggregate economy after 1985. Among these studies, Thangavelu (2004) adjusts labor input for dif- ferent skill levels and also employs capacity utilization rate for capital stock growth estimates. Hence, its findings are rather lower than others. Differing results of these studies are mainly due to construction of data and differing estimates of capital stock.
Econometric estimates of TFP growth are of two kinds. The stochastic frontier approach decomposes productivity growth into technical progress, technical efficiency change, scale economies, and allocative efficiency com- ponents. For this purpose, first a functional form of production function (such as Cobb–Douglas or translog) is specified and then the parameters
Table 6.2. Selected Previous TFP Growth Estimates of Singapore (Annual Averages in Percentage).
Time Finding
Source Methodology Period Dimension (%)
Chen (1977)∗ Conventional growth accounting
1957–1970 Aggregate economy
3.6 Ikemoto
(1986)∗
Conventional growth accounting
1970–1980 Aggregate economy
1.8 Elias (1990)∗ Conventional growth
accounting
1950–1987 Aggregate economy
1.8 World Bank
(1993)
Conventional growth accounting
1960–1989 Aggregate economy
1.7 Kawai
(1994)∗
Conventional growth accounting
1970–1980 Aggregate economy
0.7 1980–1990 Aggregate
economy
1.6 Lindauer and
Roemer (1994)∗
Conventional growth accounting
1965–1990 Aggregate economy
3.6
Rao and Lee (1995)
Conventional growth accounting
1976–1984 Manufacturing −0.4 1987–1994 Manufacturing 3.2 Collins and
Bosworth (1996)∗
Conventional growth accounting
1960–1994 Aggregate economy
1.5 1973–1984 Aggregate
economy
1.0 1984–1994 Aggregate
economy
3.1 Marti
(1996)∗
Conventional growth accounting
1970–1990 Aggregate economy
1.4 Leung (1997) Conventional growth
accounting
1983–1993 Manufacturing 2.8 1980–1985 Aggregate
economy
−0.6 Wong and
Sim (1997)
Conventional growth accounting
1985–1990 Aggregate economy
3.8
(Continued)
Table 6.2. (Continued)
Time Finding
Source Methodology Period Dimension (%)
1990–1998 Aggregate economy
1.2 1970–1985 Aggregate
economy
−2.1 Thangavelu
(2004)
Conventional growth accounting
1986–1995 Aggregate economy
1.6 1970–1995 Aggregate
economy
−0.7 1966–1972 Manufacturing 0.6 Tsao (1982) Translog production
function
1972–1980 Manufacturing −0.9 Wong and
Gan (1994)
Translog production function
1981–1990 Manufacturing 1.6 1966–1990 Aggregate
economy
−0.2 Young (1992
and 1995)
Translog production function
1970–1990 Manufacturing −1.0 Bercuson
(1995)
Translog production approach
1961–1991 Aggregate economy
1.8 Kim and Lau
(1994)
Meta production function
1960–1990 Aggregate economy
0.0 Mahadevan
and Kalirajan (2000)
Stochastic
production frontier
1976–1984 Manufacturing 0.9
1987–1994 Manufacturing −0.5 Iwataet al.
(2002)
Nonparametric derivative estimation
1960–1995 Aggregate economy
3.7
Sun (2002) Stochastic coefficients production frontier analysis
1970–1997 Manufacturing −0.8
Kohet al.
(2004)
Stochastic frontier analysis
1975–1998 Manufacturing 1.8
∗Cited from Felipe (1997).
of this function are estimated using the maximum likelihood estimation method.
As listed in Table 6.1 studies using the stochastic frontier approach produce different results for productivity growth. In addition to stochastic frontier approach, Kim and Lau (1994) estimated a so-called “meta- production function” by using panel data from East Asian and some developed countries. They found zero TFP growth for Singapore as well as for Korea and Taiwan. Another econometric approach is the non-parametric estimation by Iwataet al.(2002). This study makes no assumption on the competitive structure of the factor markets or the functional form of the pro- duction function. It uses a nonparametric derivative estimation technique to decompose output growth into its sources by estimating the output elas- ticities of capital and labor. This method produces the highest estimate for Singapore, at 3.7 percent level per annum for the 1960–1995 period, casting doubt on the findings of the previous approaches.
The translog approach takes into account the changes in the compo- sition of capital and labor inputs and includes such changes in the growth rates of capital and labor inputs. The estimation of TFP growth rate in this approach is similar to the residual calculation in the conventional growth accounting approach. This study also utilizes this approach to estimate TFP growth rate. Translog production function can also be estimated by econo- metric methods. There are four other studies comparable to this study. Tsao (1982), the oldest of these, utilized translog price and production functions to estimate TFP growth for manufacturing industries for the periods 1966–
1972 and 1972–1980. She found very low annual average manufacturing TFP growth rates, for these periods, 0.6 and −0.9 percent, respectively.
Additional sensitivity analyses also did not change the fact that technical change in this period was almost zero. Tsao’s study was later extended using the same methodology by Wong and Gan (1994). They found 1.6 percent TFP growth rate for the manufacturing sector for the period 1981–
1990. Young (1992, 1995) estimated TFP growth rates using the translog production function approach for the aggregate economy and for the man- ufacturing sector for the 25-year period 1966–1990. He found zero annual average TFP growth rate for the aggregate economy and−1 percent TFP growth for the manufacturing sector. Moreover, he compares his results of TFP growth for the aggregate economy with those he found for Hong Kong (2.3 percent for period 1961–1991), Korea (1.7 percent for period
1960–1990), and Taiwan (2.6 percent for period 1961–1991) and con- cludes that TFP growth performance of Singapore was the worst among the newly developing Asian economies. The fourth study was conducted by an IMF staff team (Bercuson, 1995). This study estimates TFP growth for the aggregate economy for the period 1961–1991, but unlike the previous three studies, does not adjust factor inputs for quality changes. Both labor and capital are assumed to be of the same quality from 1960 to 1991. Thus, the results (1.8 percent for the period 1961–1991) are most likely to overes- timate actual TFP growth as embodied technological change is included as part of the estimated TFP growth rate rather than advances in input quality.