Abstract
This 2011 Article IV Consultation—Selected Issues paper focuses on estimating potential output and the output gap and spillovers from agriculture in the case of Uruguay. It introduces additional economic information and theory to estimate potential output, shedding some light on the discussion of current monetary and fiscal policies. The objective is to take advantage of economic data to disentangle the most recent economic performance by introducing multivariate techniques. The paper also presents an overview of the labor market and pension system of Uruguay.
I. Uruguay: Estimating Potential Output and the Output Gap and the Spillovers from Agriculture1
A. Introduction
1. Potential output and the output gap are unobservable economic variables, yet they are critical for macroeconomic policymaking. In the case of fiscal policy, adequate estimates over the magnitude of the output gap help assess the structural fiscal policy stance, and make timely decisions to apply neutral or contra-cyclical policies as needed to ensure sustainable growth and help limit inflation pressures. In the case of monetary policy under inflation targeting regime frameworks, output gaps often feed the central bank’s implicit Taylor rules—helping determine the size of the needed adjustment to the monetary policy rate to keep inflation and inflation expectations on track.
2. This paper provides estimates of both potential output and the output gap for Uruguay based on a wide range of methods. The objective of the paper is to provide the authorities with an extensive set of estimates that can help them guide policy implementation, as well as a sense of how robust these are. The paper also presents estimates of the impact of the agricultural activity—a leading sector—on the rest of the economy.
3. The main findings of this study are as follows. First, there is a high degree of consistency among the different techniques applied in terms of the size and direction of the output gap. Second, the results based on univariate filters show some sensitivity to the length of the cycle assumed. Third, following the 2002/03 domestic financial crisis, Uruguay’s economy has undergone a substantial transformation, growth has accelerated, and it seems Uruguay is at a higher level of potential output. Four, despite the caveats discussed in the paper about the estimates, the consistency of the results across the different methods could contribute to guide the policy decision making process. Fifth, it seems that the spillovers from the agriculture sector to the rest of the economy are relatively moderate in most cases.
4. The rest of this paper is organized as follows: Section B discusses estimates of potential output and the output gap for Uruguay applying univariate filters. Section C introduces additional economic information and theory to estimate potential output, shedding some light into the discussion of current monetary and fiscal policies. The objective is to take advantage of economic data to disentangle the most recent economic performance by introducing multivariate techniques such as the Kalman filter, the production function, and a Structural Vector Auto-regressive Model. Section D analyses the spillover effects from agriculture to the rest of the economy. Section E concludes with some relevant inputs for policy analysis and decision making.
B. Potential Output and the Output Gap: Estimates with Univariate Methods
5. Policy makers and researches alike measure the position of a country’s economy in the business cycle based on estimates of potential output. Policy advice and decision making tend to rely heavily on unobserved measures of potential output and the output gap. Potential or trend output can be thought as the level of GDP if prices were fully flexible while the output gap represents the cyclical component of actual GDP compared to potential GDP. Potential GDP can also be thought as the level when the economy is at full employment. Furthermore, potential output can be thought as the level of GDP at its long term trend.
6. The different univariate techniques presented here rely on GDP time series to estimate the long term or permanent component (potential GDP) and its cyclical portion (the output gap) of the economy rate of growth. Two types of methods are discussed in this section: i) univariate filters, which include: a) the Hodrick-Prescott filter, b) the Baxter and King filter, and c) the Christiano-Fitzgerald filter; and ii) the piece-wise linear de-trending (Box 1).
7. The univariate filters presented here can be classified as two-sided filters. They are called two-sided because they use historical data as well as GDP estimates. The latter information is included to reduce the well known “end of the period bias”—a common statistical caveat of this approach, as estimates of potential output are heavily pulled by the most recent observations in the sample. The data used for this paper is quarterly GDP and the period covered is from Q1 1977 to Q4 2010. To reduce the end of period bias, quarterly projections based on Fund’s staff estimates for 2011-2016 are used; the projections are very much in line with consensus for this year and the medium term.
8. Results from these methods suggest that Uruguay’s potential output is in the range of 2.7 percent and 2.8 percent for the whole sample period. In addition, all methods suggest that Uruguay has a positive output gap for both 2010 and 2011, where actual output exceeds potential by 0.9 percent and 0.7 percent of potential GDP, respectively. Table 1 summarizes the result of the different methods applied:
The HP filter estimates potential output growth at 6.3 percent and 5.8 percent for 2010 and 2011, respectively. The output gap, which was positive up to 2008, turned negative in 2009 reflecting the spillovers from the global economic recession. With the strong growth recorded in 2010, the output gap became positive once again and for 2011 is estimated at 1.0 percent of potential GDP.
The Baxter-King filter, which yields the highest estimate of potential output among the univariated methods, suggests that trend growth was 6.5 percent and 6.3 percent for 2010 and 2011, respectively. For 2010, the estimated output gap was 0.6 percent and nearly closed by the end of 2011.
Under the Christiano-Fitzgerald filter, trend output growth averaged 6.1 percent for 2010-2011 with a positive output gap close to 1.0 percent in 2010 but, similar to the Baxter –King estimate, starting to close in 2011.
Uruguay: Potential Output and the Output Gap
Uruguay: Potential Output and the Output Gap
| Potential GDP Growth Rate | Output gap | |||||||
|---|---|---|---|---|---|---|---|---|
| Univariate Filters | 77Q1-11Q1 | 77Q1-02Q4 | 03Q1-11Q1 | 2010 | 2011 | 2010 | 2011 | |
| Hodrick-Prescott | 2.7 | 2.2 | 4.8 | 6.3 | 5.8 | 0.8 | 1.0 | |
| Piece-wise linear de–trending | 2.8 | 2.6 | 5.9 | 5.9 | 5.9 | 1.2 | 1.3 | |
| Baxter and King | 2.8 | 2.4 | 5.3 | 6.5 | 6.3 | 0.6 | 0.2 | |
| Christiano-Fitzgerald (1) | 2.7 | 2.2 | 5.1 | 5.6 | 6.5 | 0.9 | 0.4 | |
| Average univariate filters | 2.8 | 2.3 | 5.3 | 6.1 | 6.1 | 0.9 | 0.7 | |
| 90Q1-11Q1 | 90Q1-02Q4 | 03Q1-11Q1 | 2010 | 2011 | 2010 | 2011 | ||
| Multivariare Filters | ||||||||
| Kalman & HP | 3.2 | 2.4 | 4.8 | 6.2 | na | 0.9 | 1.0 | |
| Kalman HP & PC | 3.3 | 2.7 | 4.9 | 6.8 | na | 0.5 | na | |
| Kalman HP & OL | 3.3 | 2.7 | 4.6 | 6.5 | na | 1.3 | na | |
| Kalman HP & IS | 3.3 | 2.7 | 4.8 | 6.5 | na | 0.6 | na | |
| Average Kalman filters | 3.3 | 2.6 | 4.8 | 6.5 | na | 0.9 | na | |
| Structural VAR (B & Q method) | 3.4 | 2.6 | 5.4 | 8.0 | na | 0.2 | na | |
| Production Function Approach | 1997-2010 | 1997-2002 | 2003-2010 | 2010 | 2011 | 2010 | 2011 | |
| Growth accounting | 1.9 | -2.0 | 4.3 | 5.7 | 5.4 | 4.9 | 5.5 | |
| VECM | 2.3 | -1.8 | 4.9 | 6.2 | 5.8 | 0.8 | 1.0 | |
| Average Production Function: | 2.1 | -1.9 | 4.6 | 6.0 | 5.6 | 2.8 | 3.2 | |
Uruguay: Potential Output and the Output Gap
| Potential GDP Growth Rate | Output gap | |||||||
|---|---|---|---|---|---|---|---|---|
| Univariate Filters | 77Q1-11Q1 | 77Q1-02Q4 | 03Q1-11Q1 | 2010 | 2011 | 2010 | 2011 | |
| Hodrick-Prescott | 2.7 | 2.2 | 4.8 | 6.3 | 5.8 | 0.8 | 1.0 | |
| Piece-wise linear de–trending | 2.8 | 2.6 | 5.9 | 5.9 | 5.9 | 1.2 | 1.3 | |
| Baxter and King | 2.8 | 2.4 | 5.3 | 6.5 | 6.3 | 0.6 | 0.2 | |
| Christiano-Fitzgerald (1) | 2.7 | 2.2 | 5.1 | 5.6 | 6.5 | 0.9 | 0.4 | |
| Average univariate filters | 2.8 | 2.3 | 5.3 | 6.1 | 6.1 | 0.9 | 0.7 | |
| 90Q1-11Q1 | 90Q1-02Q4 | 03Q1-11Q1 | 2010 | 2011 | 2010 | 2011 | ||
| Multivariare Filters | ||||||||
| Kalman & HP | 3.2 | 2.4 | 4.8 | 6.2 | na | 0.9 | 1.0 | |
| Kalman HP & PC | 3.3 | 2.7 | 4.9 | 6.8 | na | 0.5 | na | |
| Kalman HP & OL | 3.3 | 2.7 | 4.6 | 6.5 | na | 1.3 | na | |
| Kalman HP & IS | 3.3 | 2.7 | 4.8 | 6.5 | na | 0.6 | na | |
| Average Kalman filters | 3.3 | 2.6 | 4.8 | 6.5 | na | 0.9 | na | |
| Structural VAR (B & Q method) | 3.4 | 2.6 | 5.4 | 8.0 | na | 0.2 | na | |
| Production Function Approach | 1997-2010 | 1997-2002 | 2003-2010 | 2010 | 2011 | 2010 | 2011 | |
| Growth accounting | 1.9 | -2.0 | 4.3 | 5.7 | 5.4 | 4.9 | 5.5 | |
| VECM | 2.3 | -1.8 | 4.9 | 6.2 | 5.8 | 0.8 | 1.0 | |
| Average Production Function: | 2.1 | -1.9 | 4.6 | 6.0 | 5.6 | 2.8 | 3.2 | |
Methods to Estimate Potential Output and the Output Gap
There is a wide range of methods to estimate potential output and the output gap. These include: a) univariate methods; b) multivariate methods; and c) economic models such as structural vector autoregressive models. This box describes the basic features of univariate and multivariate filters.
Univariate methods:
The Hodrick-Prescott filter is the most widely used technique to estimate potential output. This method estimates the trend component minimizing the deviations of actual GDP from its trend level. This is achieved imposing a trade-off between the fit of the sample data and the degree of smoothness of the estimated trend output series. In line with the standard practice for quarterly observations, λ is set at 1,600. The higher the penalty λ, the smoother the trend series become as λ reflects the maximum in change allowed in potential growth in two consecutive periods.
The Baxter and King filter is classified as a band pass filter, which removes the slow moving components (trend growth) as well as the high frequency (cyclical) elements while keeping the intermediate components (business cycle) of the GDP series. In this case, the duration of the cycle has to be defined. The standard is to assume that the cycle last between 1.5 and 8 years. If using quarterly data, then the required parameters have to be set at 6 and 32.
The Christiano-Fitzgerald filter is also a band pass filter. In the same way to the Baxter and King filter, it adjusts the business cycle for different frequencies of the cycle over the sample data of actual GDP. In this method, the business cycle is thought as fluctuations of a certain frequency.
The piece-wise linear de-trending method is a technique that can be applied to data than includes structural breaks points in the sample period. The advantage of this technique is that it considers different trends in different subsamples within the time series. In this case, it fits a linear trend through the logs of the quarterly GDP series, which has to be tested for structural breaks applying the Chow breakpoint test and the Quandt-Andrews test
Multivariate methods
The four Kalman filters presented in this paper have an advantage over univariate filters as they incorporate additional economic variables to decompose the permanent and cyclical component (state variables which are not observable) of the actual rate of growth. The Kalman filters estimate trend output and the output gap that are most consistent with observed variables such as inflation, the monetary policy rate, and the rate of unemployment.
9. Changing the sample period does seem to affect slightly the estimates of potential output. Comparisons are presented here only for the HP filter, but all the methods yield similar results. Compared to the larger sample, estimates using a subsample covering Q1 1987 to Q4 2010 –that is excluding 40 observations, generates estimates of potential output slightly lower. For 2010, potential growth was estimated at 6.1 percent while the output gap was estimated at 0.8 percent.
10. Applying a piece-wise linear de-trending method (PWLD) to measure Uruguay’s potential output is important given the impact of the 2002 financial crisis on economic activity. Its advantage over the three previous methods is that the PWLD method considers different trends in different subsamples within the GDP series. The Chow breakpoint test and the Quandt-Andrews test detect a structural breakpoint in Q 2 2002—in line with Uruguay’s financial crisis. The PWLD method indicates that before this breakpoint, the economy was growing at a potential annual average of 2.6 percent. After the crisis, potential growth has increased to an annual average of 5.9 percent. Similar to the other methods described above, there was a positive output gap in 2010, which for this method was estimated at 1.2 percent.
Uruguay: Potential Output and the Output Gap, 1977 Q1 - 2011 Q1
Citation: IMF Staff Country Reports 2011, 376; 10.5089/9781463926601.002.A001
Source: IMF staff calculations.C. Measuring Potential Output and the Output Gap with Economic Procedures
11. Policy advice and decision making based only on statistics methods should be taken with caution given the limitations of such techniques. The main weakness of the univariate methods is that their estimates of potential output are based solely on the observed GDP series.
12. Economic theory can help to overcome such limitations. Theory tells us that there is a relationship between the output gap and trends in inflation, as well as between the output gap and unemployment. Thus, estimates of potential output and the output gap can be enhanced applying economic procedures which incorporate additional economic variables to decompose the permanent and cyclical component of the actual rate of growth. The economic methods discussed in this section include: i) the Kalman filter, which builds on Fuentes et al (2007), and estimates potential output under four different models; ii) the production function; and iii) an structural vector autoregressive model based on the Blanchard and Quah method. Results are summarized in Table 1.
Kalman Filter-based models
13. Potential output and the output gap, under the Kalman filter, are estimated applying four alternative models.
Quarterly GDP HP Model. Model one is a state-space-form model based on the quarterly GDP series which approximates the HP filter. In this case, for 2010, potential output growth was 6.2 percent with an output gap of 0.9 percent, very close to the standard HP filter.
Phillips Curve. Model two is based on the Phillips curve. In this case, potential output will be estimated including in the model the observed GDP and the inflation target set by the monetary authority. If there is a positive output gap, then the observed inflation will be above the official inflation target. One caveat in the case of Uruguay is that the inflation target regime started only in 2008, so the number of observations limits the estimates. To deal with this shortcoming, we use as a proxy of the official targets, which corresponds to the inflation rates indicated in the BCU’s communiqués when the monetary authority started following the monetary aggregates back in September 2004. In this case, potential output growth in 2010 was estimated at 6.8 percent and the output gap at 0.5 percent.
Okun Law. Model three is based on Okun’s Law. Given the theoretical relationship between output and unemployment, we should expect a decline in the unemployment level beyond the natural rate of unemployment—defined here as the non accelerating inflation rate of unemployment (NAIRU), if the economy is operating above potential growth. On the contrary, we should expect an unemployment level above the NAIRU if the output gap is negative. Under this approach, in 2010 potential output grew 6.5 percent and the output gap was 1.3 percent.
IS Curve. Model four is based on the IS curve. According to economic theory, there is relationship between the output gap and the monetary policy rate. We could expect a negative output gap when the monetary policy rate is above the equilibrium interest rate. This method suggests that for 2010, trend output growth was 6.5 percent while the output gap was 0.6 percent.
Uruguay: Potential Output and the Output Gap, 1990 Q1 – 2011 Q1
Citation: IMF Staff Country Reports 2011, 376; 10.5089/9781463926601.002.A001
Source: IMF staff calculations.Production function-based model
14. An aggregate production function is estimated based on a standard growth accounting technique as well as through a vector-error-correction model (VECM). In this case, potential output or GDP is related to its inputs: capital, labor, and technology through the Cobb-Douglas production function as follows:
Where Yt is total output, K is the capital stock, L is labor, A is the technology parameter or total factor productivity (TFP), and α is the share of capital in total output. TFP is calculated as a residual from the contribution of labor and capital to real GDP growth. The latter labor and capital shares are estimated through the VECM model.
Standard growth accounting technique
15. In the standard growth accounting approach, to estimate potential output, the HP filter is applied to each series. As in the previous estimates presented in the paper, we use indices, in this case for labor, capital, and total factor productivity. Once the series have been detrended, we estimate potential output substituting trend variables in the production function and applying their contribution to growth.2 Given the limited availability of historical data on labor, the sample period covers from 1997 to 2010 and includes the Fund’s staff projections on GDP, labor, and capital.
16. This approach gives an average growth rate of potential output equal to 4.3 percent for 2003-2010. According to the estimates, potential growth in 2010 reached 5.7 percent and for 2011 is estimated at 5.4 percent. The estimated output gap for 2010 and 2011, though much larger compared to the other methods, still support the view that the economy is growing above its potential. One possible explanation for the larger rate of growth in potential output at the end of the sample period may be related to the important gains in productivity associated with FDI.3
The VEC Model
17. The VECM approach helps to overcome data constraints related to the share of capital in total output. Its advantage compared to the standard growth accounting technique is that potential output can be estimated more straightforward since potential GDP is a function of capital and labor, such that at least one cointegration relationship may exists between Y, K and/or L.
18. Data series meet the time-series properties to estimate a VEC model for the production function. In this case, the null hypothesis for the presence of a unit root could not be rejected for any of the three series, while the Johansen–Juselius cointegration test indicates the presence of at least one cointegrating relationship between output and the capital stock series.
19. Estimation of potential output growth from the VEC model is a two-step process. First, the estimated parameters for the cointegrated, long-term relationship equation between Y, K and L are substituted into (1) to obtain an actual series for the ln(A). Then, the ln(A), ln(L) and ln(K) series are smoothed out using the HP filter, and can be reintroduced in the equation to compute a final estimate for the (log) of potential output.
20. The estimated VEC Model generates similar results to the other methods for potential output growth and the output gap. Potential output growth was estimated at 6.2 percent in 2010 and 5.8 percent in 2011. The VEC Model suggests that actual output is growing above potential, at around 1.0 percent in 2011, with the gap closing by 2013.
Structural Vector Auto Regression (SVAR) approach
21. Based on the approach developed by Blanchard and Quah, potential output can be estimated with aggregate supply shocks (changes in productivity) while the output gap can be estimated through aggregate demand shocks (temporary effects). In this case, after the series have been detrended with the piece-wise linear method, a vector auto regression is estimated on GDP growth and unemployment (in levels). The impulse response generated by the VAR and the residuals are decomposed into the supply and demand shocks.
22. This is achieved imposing a “zero” long run effect from the demand side shock. Consequently, potential output is estimated by restricting the demand shock to zero while allowing the supply shocks to operate. Considering there was a structural break following the 2002 crisis, the piece wise linear detrending method is applied to separate observations (sample) before and after the crisis (Q2 2002) and two different means are thus used for the two sub-periods.
23. The structural VAR yields relatively different results to the univariate filters and the economic methods. The estimations of potential output applying the SVAR show that potential output grew on an annual average of 2.6 percent before the crisis and jumped to an annual average of 5.3 percent after the crisis. For 2010, estimated potential growth exceeded all the other methods estimates; however, the output gap was nearly closed. In addition, as Figure 4 shows, there is an increasing tightness in the labor market as the actual unemployment rate is very close to its long term trend level.
Uruguay: structural VAR: output Gap, 1991 Q1 - 2010 Q4
Citation: IMF Staff Country Reports 2011, 376; 10.5089/9781463926601.002.A001
Source: IMF staff calculations.Uruguay: Labor Market: Slackness/Tightness
(Actual minus NAIRU or trend Unemployment rate)
Citation: IMF Staff Country Reports 2011, 376; 10.5089/9781463926601.002.A001
24. In addition to the caveats already mentioned in the paper, estimates of potential output for Uruguay should be treated with caution given the 2002 financial crisis. The end of period bias in the case of univariate filters and few observations to estimate potential output using the Kalman filter and the Phillips curve are the main caveats previously mentioned in this paper. More importantly in the case of Uruguay is the impact from the 2002 financial crisis on the economy. The financial crisis caused a sharp fall in output which has been followed by seven years of very strong growth.
25. The methods applied here capture a large part of the 2002 drop in actual GDP as a fall in potential GDP followed by strong growth in potential GDP. As indicated by Rosales (2011), it is likely that the recession and the banking crisis caused a fall in potential output; however, a lot of uncertainty about their magnitude remains. One effect of the crisis and the rapid recovery is that potential growth was estimated at about 6.0 percent in 2010 when applying almost all the methods, a rate that few observers believe is Uruguay’s long-term potential growth rate.
26. Thus, in spite of the consistency in the estimates across the battery of methods, estimates of potential output should be treated with prudence. All methods indicate that the economy is growing above potential; furthermore, the continuous outperformance of consumption over GDP, the level of inflation persistently above the official target, and the tightness in the labor market, all support the estimate of a positive output gap for 2010-11. To further test the robustness of the estimates, future work could focus on calculating potential output growth in real time as in Flores and Vazquez-Ahued (2011). Caution should be exercised even in this case as data revisions usually have an important effect on the estimates.
D. Spillovers from the Agriculture Sector to the Rest of the Economy
27. Following the 2002 crisis and supported by favorable prices in the international markets, agriculture production has increased by an annual average of nearly 3 percent. Soya production in Uruguay has tripled since 2005 with the cultivated area increasing from around 300 thousand hectares in 2005 to around 1 million hectares in 2010 making this one of the most dynamic sectors of the economy. Spillovers from agriculture to the rest of the economy will be assessed through a Vector Auto Regressive (VAR) model. Impact from agriculture to the other sectors will be tested through impulse response functions (IRF) and a forecast error variance decomposition analysis. The impact from agriculture to the rest of the sectors is modeled following the work by Acosta (2011). In this case, we assume the following model:
where A(L) and B(L) are a n x n and a n x k polynomial matrices in the lag operator L, respectively, Yt is a n x l vector of endogenous variables, Xt is a k x l vector of exogenous variables, and Ut is a n x l vector of estimated residuals.
28. The VAR model is specified with agriculture (Agriculture) as the most exogenous sector. It is then followed by the industrial (Industryt), construction (Constructiont), and services sectors (Servicest). This order seeks to show that a shock to the agricultural sector at period t have a contemporaneous effect on the rest of the sectors included in the model; on the other hand, a shock at time t to the other sectors will affect the rest of the sectors included in the model only with a lag. As in the case of Acosta (2011), the Dummyt variable in vector Xt controls for the severe drought that affected the economy, especially agriculture, during 2009. The model is specified as follows:
and
29. The sample period included in the VAR model is from 1997Q1 to 2011 Q1. The economic sectors data included in Yt is at constant prices, is seasonally adjusted, and is expressed in logs. Following the standard techniques and building on Acosta’s work, to estimate the VAR model, the paper uses 2 lags.
30. The results from the VAR exercise seem to indicate that the spillovers from agriculture to the other sectors of the economy are relatively limited. The impulse-response functions presented in Figures 5 indicate that a 10 percent increase in agriculture activity affects the service sector by around 1 percent, but its effect fades away almost after two quarters. In the case of industrial production, it increases near 2 percent, but its effect turns negative after the first quarter and with some positive impact again after the fourth quarter. In the case of construction, there is a temporary increase in the sectors’ activity also close to 2 percent, but its effect turns negative after one quarter more than offsetting the initial positive impact, but such negative performance vanishes after the fourth quarter.
Uruguay: Impulse-Response Functions to a 10 Percent Increase in Agricultural Output
(In percent)
Citation: IMF Staff Country Reports 2011, 376; 10.5089/9781463926601.002.A001
Source: IMF staff calculations.31. The forecast error decomposition analysis indicates that agriculture has some medium impact on the other sectors of the economy. This analysis yields information about the relative importance of an agriculture shock into the rest of the economic sectors. In the most extreme case, a shock to agriculture explains more than 50 percent of the change in activity in the services sector, but as Table 2 shows, its effects fades away faster than in the other sectors. In the case of industrial production, a shock to agriculture seems to explain one quarter of the sector’s changes in economic activity. Though its effects is also short lived, it fades away slower compared to the services sector.
Forecast Error Variace Decomposition Due to an Agricultural Sector Shock
(In percent)
Forecast Error Variace Decomposition Due to an Agricultural Sector Shock
(In percent)
| Quarters | Industry | Construction | Services |
|---|---|---|---|
| 1 | 25.7 | 34.9 | 53.5 |
| 4 | 15.1 | 27.9 | 22.8 |
| 8 | 12.6 | 23.8 | 17.7 |
| 10 | 11.5 | 22.2 | 16.1 |
Forecast Error Variace Decomposition Due to an Agricultural Sector Shock
(In percent)
| Quarters | Industry | Construction | Services |
|---|---|---|---|
| 1 | 25.7 | 34.9 | 53.5 |
| 4 | 15.1 | 27.9 | 22.8 |
| 8 | 12.6 | 23.8 | 17.7 |
| 10 | 11.5 | 22.2 | 16.1 |
32. To supplement the previous VAR analysis, a set of individual regressions are estimated. These regressions can be depicted in the following form:
where each sector’s output –other than agriculture, is represented by Yj,t; At is agricultural output, and Dummyt is the dummy variable, as in the VAR analysis, that controls for the 2009 drought that negatively affected Uruguay, especially agriculture. Quarterly data, seasonally adjusted, and in logs terms is used, and the sample period included is from 1997Q1 to 2011Q1. In these regressions, the short-run impact of agriculture on the other sectors is given by c(3) and the long-run impact by
OLS Estimation of Spillover Effects From the Agricultural Sector
OLS Estimation of Spillover Effects From the Agricultural Sector
| c(1) | c(2) | c(5) | c(4) | c(3) | Short-run effect | Long-run effect | R-squared | |
|---|---|---|---|---|---|---|---|---|
| Industry | 1.02 | -0.95 | 0.13 | 0.14 | -0.02 | 0.13 | 0.14 | 0.90 |
| t-Statistic | 0.81 | -15.69 | 2.22 | 0.06 | -0.49 | |||
| Construction | 5.61 | -0.82 | 0.34 | 0.57 | -0.03 | 0.34 | 0.50 | 0.74 |
| t-Statistic | 2.72 | -10.19 | 3.60 | 5.18 | -0.68 | |||
| Services | 0.02 | -1.03 | 0.14 | 0.18 | 0.02 | 0.14 | 0.16 | 0.97 |
| t-Statistic | 0.04 | -29.72 | 6.67 | 7.81 | 0.04 |
OLS Estimation of Spillover Effects From the Agricultural Sector
| c(1) | c(2) | c(5) | c(4) | c(3) | Short-run effect | Long-run effect | R-squared | |
|---|---|---|---|---|---|---|---|---|
| Industry | 1.02 | -0.95 | 0.13 | 0.14 | -0.02 | 0.13 | 0.14 | 0.90 |
| t-Statistic | 0.81 | -15.69 | 2.22 | 0.06 | -0.49 | |||
| Construction | 5.61 | -0.82 | 0.34 | 0.57 | -0.03 | 0.34 | 0.50 | 0.74 |
| t-Statistic | 2.72 | -10.19 | 3.60 | 5.18 | -0.68 | |||
| Services | 0.02 | -1.03 | 0.14 | 0.18 | 0.02 | 0.14 | 0.16 | 0.97 |
| t-Statistic | 0.04 | -29.72 | 6.67 | 7.81 | 0.04 |
33. The individual OLS regressions confirm there are moderate spillovers effects from agriculture to the rest of the economy. Similar to the impulse response functions, industry and services reflect the smallest impact from agriculture; meanwhile, agriculture’s impact on construction is as much as twice that of industry and services.
34. In addition to the VAR and the OLS regressions, we estimate a set of rolling VARs to assess the impact over time of agriculture to the rest of the economy. In this case, the set of rolling VARs includes 29-window period, which begins in 1997Q1. The impulse response functions to one standard deviation shock to the agriculture sector are estimated and their corresponding one-quarter response of each sector is stored. For each subsequent sample, one quarter is added to the previous sample and the earliest observation is dropped. The last sample goes from 2004Q2 to 2011Q1. Figure 6 shows the one-quarter responses from each sector. As previously indicated, all the sectors show somehow similar sensitivities to the agricultural sector.
Uruguay: Rolling VARs
(In percent)
Citation: IMF Staff Country Reports 2011, 376; 10.5089/9781463926601.002.A001
Source: IMF staff calculations.E. Conclusions
35. Potential output and the output are two key variables for fiscal and monetary policy. However, prudence should be exercised upon relying on a single approach to estimate potential growth as there are particular limitations to certain methods; thus individual results should be treated with some caution. Future work to improve potential output calculations could focus on real time estimates.
36. This paper covered a wide range of methods to estimate potential output. All of the methods indicate that potential output has accelerated following Uruguay’s 2002 financial crisis rising at an annual average of 5 percent for all the methods presented with the four Kalman Filters estimating that potential output averaged around 4.8 percent after the 2002 crisis. Among the several methods presented here, the Christiano-Fitzgerald filter and the Production Function based on a growth accounting technique yield the lowest potential growth rate for 2010 equal to 5.6 and 5.7 percent, respectively.
37. The positive output gaps that all methods generate seem to indicate that the economy remains growing above trend. With inflation and inflation expectations above the official target range, with consumption outperforming economic growth, and unemployment at historical lows and tightness in some sectors of the economy, estimating potential output and the output gap remain critical for policy decision making.
38. The spillovers from agriculture to the other sectors of the economy are moderate. A 10 percent increase in agriculture output leads to relatively small increases in the other sectors with the effect fading mostly after two to four quarters.
References
Blanchard, O. and D. Quah, 1989, “The Dynamic Effect of Aggregate Supply and Demand Disturbances,” The American Economic Review, Vol. 79, pp. 655-73.
Bucacos, Elizabeth, 2001, “Tendencia y ciclo del producto uruguayo”, Revista de Economía, Vol. 8, No. 2, Segunda época, Banco Central del Uruguay.
Acosta-Ormachea, Santiago, 2011, “Paraguay: Selected Issues Paper” IMF Country Report No. 11/189, pp. 3-15 (Washington: International Monetary Fund).
Flores, E. and F. Vazquez-Ahued, 2011, “Mexico: Selected Issues Paper” IMF Country Report No. 11/183, pp. 2-12 (Washington: International Monetary Fund).
Fuentes, R., F. Gredig, and M. Larrain, “Estimating the Output gap for Chile”, Central Bank of Chile Working Papers No. 455 (Santiago: Central Bank of Chile).
Loayza, N., P. Fajnzylber, and C. Calderon, 2004, “Economic Growth in Latin America and the Caribbean: Stylized Facts, Explanations and Forecasts”, Central Bank of Chile Working Papers No. 265 (Santiago: Central Bank of Chile).
Magud, N. and L. Medina, 2010, “Chile: Selected Issues Paper” IMF Country Report No. 10/299, pp. 51-57 (Washington: International Monetary Fund).
Theoduloz, Tania, 2005, “El producto potencial en la Economía Uruguaya: 1978-2003.”, XX Jornadas Anuales de Economía, Montevideo: Banco Central del Uruguay.
Prepared by Manuel Rosales Torres.
Following Bucacos (2001) who found a capital participation equivalent to 0.32 and Theoduloz who estimates capital participation at 0.27, in the paper we assume an α equal to 0.3 percent. Labor force data comes from the Instituto Nacional de Estadistica de Uruguay while data on capital comes from national accounts.
In addition, the increase in the TFP may be capturing some of the impact from the positive commodity export prices in some key sectors that are not reflected in the stock of physical and human capital, and that generate incentives for increased productivity.
