Potential Output Estimates—A New Look1
Vietnam has undergone a major transition in the last decade. This paper seeks to assess how this transformation has affected its growth potential. Employing a range of methodologies, the analysis concludes that Vietnam’s medium-term growth potential has increased from 6.2 percent estimated in 2014 to 6.5 percent. Acceleration of reforms that have generated productivity gains in the last decade, including the implementation of agreed free trade agreements, could further boost growth potential.
1. Vietnam has undergone a major transformation in the last decade. A very productive export-oriented sector financed by FDI has boomed, and in 2017, was responsible for more than two-thirds of Vietnamese exports and a third of the ASEAN’s tech exports. Urbanization has progressed and, concomitantly, employment has continued to shift away from agriculture toward industry and the service sectors, further boosting productivity. The large state-owned enterprise (SOE) sector, which has historically been a barrier to the development of a vibrant private sector, is undergoing equitization and reforms. These measures are helping to limit state involvement in the economy by restricting SOE operations to core areas. Together, these forces have contributed to increase Vietnam’s productivity.
2. This paper seeks to understand how Vietnam’s transition has affected its growth potential. The most recent IMF estimates of potential GDP for Vietnam date back to 2014, before the FDI-boom started in earnest.2 Vietnamese authorities do not publish potential output calculations. The empirical foundations for estimating potential output have also evolved following the GFC. This paper uses several approaches—both well-established in the literature and new methodologies—to estimate potential output for Vietnam: (i) a Hodrick-Prescott (HP) filter; (ii) a multivariate filter; (iii) a multivariate filter augmented by financial frictions; and (iv) a production function approach. Appendix I contains the details on each methodology.
A. Potential Output Estimates Using A HP Filter
3. The HP filter, a purely statistical tool used to extract a trend component from a time series, is applied to GDP. The trend is chosen to minimize a loss function that depends on both the deviation of the trend from actual GDP and the curvature of the trend. The smoothing parameter λ determines the relative weight of these two objectives. While the HP filter’s main qualities are its simplicity and transparency, it remains a purely statistical tool and lacks economic relevance. In addition, the filter suffers from the well-known end-point bias, because the last point of the series has an exaggerated impact on the trend. To solve this problem and obtain a more robust estimate at end-2017, we complement the GDP time series with staff GDP projections for the period 2018Q1–2019Q4. With this approach, potential growth is estimated at 6.5 percent in 2017 and the output gap at 0.75 percent (text Figure).
Vietnam: HP Filter, 2000Q1–2018Q4
Sources: Authorities’ Data and IMF Staff estimates.
B. Potential Output Estimates Using A Multivariate Filter
4. The multivariate filter approach (MVF) estimates potential GDP based on a model that captures relationships between actual and potential GDP, inflation and unemployment (Blagrave and others, 2015). Potential output is derived as the level of output that can be achieved when inflation and unemployment are at their equilibrium levels. Bayesian techniques are used to estimate a system of equations that describe the evolution of the 3 key observable variables (output, core inflation and unemployment) and their relationship with potential output (output gap, Phillip’s curve and Okun’s law). The model’s equations and calibration are detailed in Appendix I. Data on growth and inflation expectations (WEO projections for past vintages) are added, in part to help identify shocks, but mostly to improve the accuracy of estimates at the end of the sample period.
5. Using the MVF approach, potential growth is estimated at 6.4 percent in 2017 and the output gap at -0.1 percent. While, unlike the HP filter, this methodology relies on economic intuition, there are several shortcomings to applying it to an economy like Vietnam that are worth mentioning. First, as described in Box 1, the relationship between the output gap and inflation is weak. Furthermore, the official unemployment data series exhibits very little variation over the estimation period and unemployment does not seem to respond to aggregate demand shocks. (for example, unemployment was higher in 2004–07—when the output gap was positive—than in 2011–16, after the country was hit by a domestic financial crisis and a negative output gap opened up). This can be explained by the high share of the population employed in the informal, agriculture and the public sectors which are maybe less affected by cyclical conditions.
Vietnam: MVF, 1999–2020
Sources: Vietnamese authorities; and IMF Staff estimates.
Box 1.Vietnam’s Phillip’s Curve
The hybrid New-Keynesian Phillip’s curve defines the relationship between inflation, expected and lagged inflation, the lagged output gap and the exchange rate:
πt = the annualized quarterly change in the core CPI
π4t = the four-quarter change in the core CPI
yt = the output gap (calculated by HP Filter)
St = the Nominal Effective Exchange Rate
This equation is based on the idea that the role of monetary policy is to provide a nominal anchor for inflation, while pursuing other objectives, such as output stabilization and, in the case of Vietnam, exchange rate stability. It describes the tradeoffs that monetary authority faces between its different objectives. In this definition, some firms (αnld) optimally adjust their prices by taking into account expected inflation while the remaining firms (1 – αnld) either don’t adjust their prices or adjust based on the most recently observed rate of inflation. Thus, the behavior of the economy depends critically on the value of αnld. If it is close to 1, a small monetary policy action (e.g. increase of interest rates) will have a large and rapid effect on inflation and consequently a limited negative impact on growth. The parameter αnld can also give some insight on the credibility of the monetary policy. The literature shows that economies with weak monetary credibility have low levels of forward-looking expectations as the inflation target is less likely to anchor expectations.1 This feature is found in highly dollarized economies (such as Vietnam historically), worsening tradeoffs between output and inflation stabilization and increasing the cost of carrying out a stabilization process.2
Sources: Vietnamese authorities; and IMF Staff estimates.
αz measures the exchange rate pass-through to core inflation while αGAP measures the effect of the output gap, that is the excess/slack of aggregate demand, on inflation.
To analyze the effect of these variables on Vietnam’s inflation, we conduct a regression analysis to determine the parameters of the reduced-form Phillip’s curve. The coefficients on lagged and expected inflation are constrained so that they sum to one, to follow the definition given by equation (i). Finally, we use core inflation, which excludes raw food, energy and administered prices. The latter have been the main contributors to headline inflation in recent years, as the Vietnamese authorities have raised healthcare and education fees to increase cost recovery. Since the variation of these prices depends on fiscal and social policies rather than monetary policy, we exclude these prices from our analysis.
The results from the regressions are presented in Table 1. Column (1) shows the results for the whole sample period 2000–17. The coefficient on expected inflation is relatively small, indicating backward-looking expectations. The results show a large impact of the output gap on inflation as well as a small passthrough from the exchange rate. The adjusted R2 stands at 0.56 and the residuals are particularly large around the periods of high inflation.
Over the period 2013–17, core inflation has been more stable. The regression results are shown in column (2). The negative sign of the output gap coefficient is counter-intuitive, pointing to the data issues discussed earlier. The coefficient on expected inflation is greater than for 2000–17, indicating that the expectations have become more forward-looking. It also implies that the effects of the exchange rate and output gap on inflation have become less persistent while the exchange rate pass-through effect has declined further, in line with the strong stability of the exchange rate during this period. These results may also reflect to some extent credibility gains from the central bank after a period of successful macro-stabilization, also evident in the e observed de-dollarization underway in Vietnam.
|απld||0.43 ***||0.57 ***|
|1 – απld||0.57||0.43|
C. Potential Output Estimates Using a Multivariate Filter with Financial Frictions
6. This methodology deviates from the MFV in that it focuses on sustainable output levels, rather than potential output (Berger et al., 2015). The sustainable output level is the level of GDP that an economy can sustainably produce over the medium-term in the absence of imbalances. GDP can be at potential but may not be sustainable in the event of a credit boom that temporarily lifts output growth but affects inflation with a lag. This methodology maybe particularly relevant for Vietnam, which has experienced several years of sustained high credit growth. Financial cycles are incorporated in the model using two financial variables, the credit gap (calculated following BIS methodology) and the demeaned asset market price growth (excluding real estate market developments due to the lack of official real estate data for Vietnam).
Vietnam: MVF-FIN, 2005–2020
Sources: Authorities’ Data and IMF Staff estimates
7. Sustainable output growth is estimated at 6.4 percent in 2017, marginally lower than with other methods. This is consistent with the selected issues paper “Vietnam: Credit Growth and Asset Market Valuations”, which finds that asset price and credit growth appear to be stronger than warranted by fundamentals. However, the methodology contains some weaknesses that could be leading to an underestimation of the size of the financial imbalances and therefore an overestimation of potential output by this model. 3
D. Potential Output Estimates Using a Production Function Approach
8. In this approach, we use the following human-capital augmented production function:4
where: Yt = total GDP in year t; At = total factor productivity in year t; Kt = capital stock in year t; αt = income share of capital in year t; Lt = labor force in year t; Ht = human capital in year t; ht = level of human capital per unit of labor in year t; φ = return to education; and St = average years of schooling in year t
9. Determining elasticities. Due to the unavailability of Vietnam-specific values, the output elasticity with respect to labor, or income share of labor, is determined on the basis of regional estimates. To the extent that Vietnam is integrated into the Asian value chain because of the growing FDI flows from the region, one can expect Vietnam’s output elasticity trends to follow that of its predecessors in the value chain. ASEAN-4 countries (Indonesia, Philippines, Thailand and Malaysia), China (and India) show similar trends of declining income share of labor associated with rising per capita incomes. These countries also exhibit steady increases in their capital intensity (i.e. capital to labor ratio), which was identified as a main factor in the decline of labor income shares in emerging markets.5 By using the approximated linear relationship between the capital to labor ratio and the income share of labor, we estimate Vietnam’s income share of labor.
10. Human capital development. In this model, physical labor accumulation is augmented by human capital formation using data on years of schooling and return to education from the Barro-Lee database and Psacharopoulos and Patrinos (2004). Since the last reported data on schooling is as of 2010, we use a linear trend to extrapolate years of schooling in 2011–2023. It is worth mentioning that the present methodology only accounts for the “quantity” of education while the recent literature emphasizes the “quality” of education which affects the returns to education. This is particularly relevant for Vietnam, given its strong 2015 PISA scores.
Vietnam: Labor Force, 1990–2023
Sources: Vietnamese authorities; PWT 8.1; and IMF Staff estimates.
11. Labor statistics. Vietnam is a young country (median age of 26) and enjoys a substantial demographic dividend.6 However, a declining population has led to a decline in the growth of the labor force since the mid-1990s. Participation rates, male and female, have been high at around 80 percent, largely above regional averages.
Vietnam: Physical Capital, 1990–2023
Sources: Vietnamese authorities; PWT 8.1; and IMF Staff estimates.
12. Physical capital. After two decades of steady decline, investment growth picked up in 2014. In 2017, total investment grew by 10 percent. As a result, the physical capital stock growth accelerated in the last 5 years, to about 8 percent in 2017 (this includes the growth of high quality FDI capital in recent years).
Vietnam: Total Factor Productivity Growth
Sources: Vietnamese authorities; PWT 8.1; and IMF Staff estimates.
13. Total Factor Productivity. There have been three phases of productivity growth since the doi moi reforms launched in 1986. After a period of volatility, TFP growth averaged 2.1 percent until the Asian Financial Crisis. Between 2000–07, TFP grew more slowly, a little under 1 percent on average. The GFC and domestic financial turmoil in 2008 and 2011 led to a slowdown of TFP growth. Since 2013, TFP has recovered and grew at an average of 1.7 percent per year. 7 Regression analysis shows that the main drivers of TFP growth have been increasing FDI inflows, the declining share of employment in agriculture, declining credit to SOEs and increasing domestic private investment.
14. Using the augmented production function approach, we estimate potential GDP growth at 6.5 percent in 2017. Potential GDP is obtained by summing the growth rates of physical capital, human capital and labor, since the cyclical components such as capacity utilization and unemployment are not included in these series. However, the TFP growth series, which contains de facto these cyclical factors, is filtered using a HP filter to obtain a potential TFP growth.
15. The model can also be used to forecast potential output. The main assumptions regarding the GDP components are as follows.8 Investment is projected to grow at a rate of 10 percent in average, labor force growth to continue to decline at the rate of population growth and human capital growth to remain constant at 1.02 percent. We project a moderate improvement in average TFP growth due to the continued shift of labor away from agriculture and towards the more productive manufacturing sector, and increased FDI inflows and private sector participation. Regressions analysis supports a wide range of average TFP growth between 1.7 to 2.5 percent. An average TFP growth over 2 percent per year is unlikely to materialize in the current global and regional context, in which most countries have seen a worsening of productivity growth in the aftermath of the GFC and given the impact of aging on TFP growth.9 Hence, assuming continuity in the pace of structural reforms and improved fundamentals that have produced TFP growth in the recent past, we assume TFP to grow by an average 1.8 percent, similar to the 2014–17 average.
16. Given these assumptions, potential growth could average 6.8 percent over the medium-term, reaching 7.3 percent by 2023 (see text Table and Figure).
|2018–202||6.80||3.97||0.53||0.49||1.80|Vietnam: Contribution to Growth
Sources: Authorities’ Data, PWT 8.1, and IMF Staff estimates
17. The four methodologies provide a range of estimates for Vietnam’s potential output. On balance, we assess the potential growth estimate in Vietnam to be at 6.5 percent in 2017, higher than previous staff estimates of 6.2. The output gap is estimated at 0.4 percent in 2017. Potential output growth is estimated to remain at 6.5 percent over the medium-term, and the output gap is expected to close in 2019. The higher rate of potential growth can be explained by relatively high investment levels, a large and well-educated labor force which is moving towards higher value-added industries. In addition, reforms reduced the size of the SOE sector and boosted the private sector’s participation in the economy have enhanced productivity. The booming FDI sector provides an excellent opportunity to enhance the quality of capital and facilitate the transfer of technology and expertise to the domestic sector.
Vietnam: Potential Growth Estimates, 2001–20
Sources: Authorities’ Data and IMF Staff estimates.
18. This analysis will be extended further in a forthcoming paper. The production function estimates can be further improved by explicitly incorporating the effect of structural transformation due to labor reallocation into the model, and by better accounting for the impact of the quality of human capital accumulation by taking the quality of education into account. Improvements in data quality, for example, on real estate prices, quarterly GDP, unemployment rate and labor force in the informal sector, and capacity utilization, could further enhance the analysis.
1. HP filter
The standard quarterly λ for economic variables of 1,600 is used here. The GDP series on which the HP filter is applied consists of actual data for 2000Q1–2017Q4 and staff GDP projections for the period 2018Q1–2019Q4 (to minimize the end-point bias).
II. Multivariate Filter
The equations describe the evolution of the output process and its relationship with inflation and unemployment.
The level of potential output
2. Phillip’s Curve
where π is the core inflation and y the output gap.
3. Okun’s Law
where is the unemployment gap, that is the difference between the NAIRU
In addition, data on growth and inflation expectations (WEO projections for past vintages) are added, in part to help identify shocks, but mostly to improve the accuracy of estimates at the end of the sample period.
III. Phillips Curve (Box 1)
In regression (1), the statistical significance of the coefficients on output gap and exchange rate is weak, at 0.2. The estimated coefficient of the output gap could be affected by the quality of the quarterly GDP series which undermines the HP filter calculation. The exchange rate has been maintained within a tight range to the U.S. Dollar during most of the period, limiting the information available to explain changes in inflation.
A variation of the regression model, where the change in reserves and lagged credit growth are added as independent variables to the regression, increases the adjusted R2 increased to 0.74 for the period 2008–12. Reserve intervention has indeed been a major monetary policy instrument used by the SBV to limit exchange rate depreciation and as such, have impacted inflation during this period. One caveat about regression (2) results is that the small number of observations might impact negatively the robustness of the results.
IV. Multivariate Filter with Financial Frictions
Sustainable GDP is estimated by decomposing observed GDP time series into two unobservable components:
In addition, the model takes into account information from a set of observable variables, xt which could be correlated to the output gap:
where the variance ratio
V. Potential Output Estimates Using A Production Function Approach
Assumptions made for potential output forecast
Labor statistics. The participation rate is projected to remain constant during projection period, while labor force growth is projected to continue to decline at the rate of population growth, based on the United Nation’s population projections. Given the large share of informal and self-employment in Vietnam, as well as the lack of reliable comprehensive data, the unemployment rate is assumed unchanged over the period.
Human Capital. Schooling for 2011–23 is extrapolated from a linear trend. Return to education is assume constant at 7% over the period.
Physical Capital. Data on capital stock is available from the Penn World Table until 2014. For 2015- 23, the new stock is calculated using the formula:
where: Kt = Capital stock in year t; δt = Depreciation ratio in year t; It = Investment in year t.
The depreciation ratio is projected to remain constant, at 4.5 percent, while the data on investment is based on IMF baseline projections for gross fixed capital formation (for 2015–17, the data is provided by the Vietnamese authorities). In 2018–23, investment is projected to grow by 10 percent, in line with recent average growth rates. Capital stock is expected to continue to grow after sustained deceleration between 1994 and 2013. Given the unavailability of data, capacity utilization is assumed to be unchanged over the period. The actual changes of the capacity utilization are captured in the TFP.
ArgovEpsteinKaramLaxton andRose2007. “Endogenous Monetary Policy Credibility in a Small Macro Model of Israel.” IMF Working Paper 07/207International Monetary FundWashington, DC.
BenesClintonGarcia-SaltosJohnsonLaxtonManchev andMatheson2010. “Estimating Potential Output with a Multivariate Filter.” IMF Working Paper 10/285International Monetary FundWashington, DC.
BergerDowlingLanauLianMrkaicRabanal andSanjani2015. “Steady as She Goes—Estimating Potential Output During Financial “Booms and Busts”.” IMF Working Paper 15/233International Monetary FundWashington, DC.
BlagraveGarcia-SaltosLaxton and Zhang2015. “A Simple Multivariate Filter for Estimating Potential Output.” IMF Working Paper 15/79International Monetary FundWashington, DC.
Cespedes and Soto2006. “Credibility and Inflation Targeting in Chile.” Central Bank of Chile Working Papers N° 408
Dizioli and Schmittmann2015. “A Macro-Model Approach to Monetary Policy Analysis and Forecasting for Vietnam.” IMF Working Paper 15/273International Monetary FundWashington, DC.
Felices and Tuesdta2007. “Monetary Policy in a Dual Currency Environment.” Central Reserve Bank of PeruWorking Paper series April 2007
International Monetary Fund2015. “Where are we Headed? Perspectives on Potential Output.” World Economic OutlookAprilWashington, DC.
International Monetary Fund2016. “Global Disinflation in an Era of Constrained Monetary Policy.” World Economic OutlookAprilWashington, DC.
International Monetary Fund2017a. “Vietnam, Selected Issues, Demographics.” IMF Country Report No. 17/191International Monetary FundWashington, DC.
International Monetary Fund2017b. “The “new mediocre” and the Outlook for Productivity in Asia.” Regional Economic OutlookAprilWashington, DC.
International Monetary Fund2017c. “Understanding the Downward Trend in Labor Income Shares.” World Economic OutlookAprilWashington, DC.
Laxton and N’Daye2002. Monetary Policy Credibility and the unemployment-inflation trade-offIMF Working Paper 02/220International Monetary FundWashington, DC.
Maliszewski2010. “Vietnam: Bayesian Estimation of Output Gap.” IMF Working Paper 10/149International Monetary FundWashington, DC.
PodpieraRaei andStepanyan2017. “A Fresh Look at Potential Output in Central, Eastern, and Southeastern European Countries.” IMF Working Paper 17/37International Monetary FundWashington, DC.
Psacharopoulos and Patrinos2014. “Returns to Investment in Education: A Further Update.” Education Economics Vol. 12 No. 2
Prepared by David Corvino (APD). I would like to thank Mr. Nguyen Duc Trung, Deputy Director General of the Forecasting and Statistics Department of the SBV for organizing the seminar as well as all the participants for their thoughtful comments.
In 2017, FDI inflows reached US$17.5 billion, up 40 percent compared to 2014.
Credit gap is calculated using a HP filter and therefore, it is subject to end-point bias. Moreover, the estimates of credit gap may also be affected by weaknesses in GDP and credit data quality and by the fact that the estimates do not incorporate the effect of structural factors underlying the economic transformation underway in Vietnam.
The assumptions made for estimating this model are detailed in Appendix I.
Due to data constraints, TFP also captures other factors such as education quality, capacity utilization and average hours worked.
Detailed assumptions are described in Appendix 1.