A. Introduction

1. Colombia’s real effective exchange rate has fluctuated considerably in recent years. After reaching its most depreciated level in at least 50 years in early 2003, the real exchange rate appreciated by 33 percent between March 2003 and March 2006. In April 2006, market sentiment changed abruptly and by June, the real exchange rate had depreciated again by 14 percent, although most of that decline reversed again over the following two months (Figure 1).³²

Real Effective Exchange Rate, 1996/1-2006/8

(Index: 2000=100)

Citation: IMF Staff Country Reports 2006, 401; 10.5089/9781451808919.002.A004

• Download Figure
• Download figure as PowerPoint slide

View Full Size

Real Effective Exchange Rate, 1996/1-2006/8

(Index: 2000=100)

Citation: IMF Staff Country Reports 2006, 401; 10.5089/9781451808919.002.A004

• Download Figure
• Download figure as PowerPoint slide

Real Effective Exchange Rate, 1996/1-2006/8

(Index: 2000=100)

Citation: IMF Staff Country Reports 2006, 401; 10.5089/9781451808919.002.A004

• Download Figure
• Download figure as PowerPoint slide

2. This paper aims to identify fundamental determinants of the real exchange rate and assess how the current level of the real exchange rate relates to those fundamentals. More specifically, a statistical model is estimated that links Colombia’s real exchange rate with potential fundamental determinants in order to answer the following questions:

• What are the fundamental determinants of Colombia’s equilibrium real exchange rate?

• To what extent can recent movements be justified by the behavior of economic fundamentals, or to what extent do they represent deviations from those fundamentals?

• Is the current level of the real exchange rate in line with historic patterns, and does it appear to be sustainable based on current and projected future fundamentals?

3. The remainder of the paper is organized as follows. Section II provides a brief review of the literature. Section III describes the empirical framework, including the data and the main methodology. Section IV discusses the results of various alternative specification and compares them to previous work. Section V evaluates the model results and discusses its implications. Section VI briefly presents the results of some alternative methods and considerations for assessing the real exchange rate and Section VII concludes.

B. Brief Review of the Literature

4. The literature on assessing real exchange rates, including estimating empirical models, is extensive but a clear consensus on the appropriate framework and the choice of fundamentals has not yet emerged.³³ After initially focusing on the traditional Purchasing Power Parity (PPP) hypothesis, the literature has shifted towards estimating the link between the RER and its fundamental determinants. According to PPP, the real exchange rate should remain constant in the long run, as arbitrage in international goods markets leads to price convergence. PPP seems to hold in the long run for developed countries (Froot and Rogoff, 1996; and Rogoff, 1996), but deviations can be large and convergence to the PPP levels very slow, leading to low power of statistical tests. Moreover, the logic of price equalization through arbitrage only applies strictly to individual goods that are tradable internationally, but not for broader indices that include non-tradable goods and services. Most of the recent work thus considers the possibility of non-constant levels of the equilibrium real exchange rate and aims at estimating the long run co-movements between so-called fundamentals and the real exchange rate.

5. Variables are considered fundamentals if a permanent change in the level of the variable causes a permanent change in the real exchange rate. In contrast, variables that only have temporary effects on the real exchange rate are not considered fundamentals in this context. The theoretical justification for using certain variables can be illustrated in a simple neoclassical framework with tradable and non-tradable goods that are not perfect substitutes. The model assumes price equalization for tradable goods across countries and wage equalization across sectors within countries. If the country were small relative to the rest of the world, then it would face a highly elastic supply of imported tradables from the rest of the world and also a highly elastic demand for domestically produced tradables. Demand and supply for non-tradables, on the other hand, would be more inelastic. In such a model, changes in relative demand or relative supply of tradables vs. non-tradables will affect their relative price. With price equalization of tradable goods across countries, such movements in the relative price of tradables and non-tradables cause movements in the real exchange rate and thus explain deviations from PPP. The most important variables that have been used in the literature are:³⁴

• Productivity differential, or Balassa-Samuelson effect (see Balassa, 1964; and Samuelson, 1964). An increase in productivity in the tradable sector would lead to rising wages in that sector to match the higher marginal product of labor. Wage equalization would then also require higher wages in the non-tradable sector, which implies higher prices of non-tradables to match the increased marginal cost. In equilibrium, the result would be a higher relative price of non-tradables and thus a RER appreciation, assuming no changes in trading partners’ economies. The crucial variable is relative productivity in the tradable vs. non-tradable sectors compared to trading partners. Since high-quality productivity data is often difficult to come by for developing countries, various proxies have been used in the literature. Productivity gains are often stronger in the tradable sector, which can help explain the persistent trend of real exchange rate appreciation often observed in countries with sustained high growth.

• Terms of trade or commodity prices. Higher commodity prices—or improvements in the terms of trade more generally—would tend to appreciate the real exchange rate through two channels. As in the case of productivity growth, higher export commodity prices would induce higher wages in both sectors and thus a higher price for non-tradables. In addition, higher export prices represent a positive wealth effect in the exporting country which would raise demand, including demand for non-tradables. Given the relatively inelastic supply of non-tradables, prices would have to rise to clear the market. The effect of improving terms of trade can also be illustrated by looking at the current account. Higher export prices would imply a higher current account balance. In order for the current account to move back in line with desired intertemporal consumption and investment, the real exchange rate would have to appreciate, stimulating imports while curtailing exports. Theoretically, the terms of trade should be the better measure, but many studies have achieved better results using commodity prices. However, for countries with a large share of commodity exports commodity price movements usually dominate the terms of trade and the two measures are often very similar. In Colombia, commodity exports still account for about half of total exports, although the share has declined substantially over the past 40 years.

• Net foreign asset position. The net foreign asset position could also affect the real exchange rate through two channels. Higher net foreign assets imply a positive wealth effect that should raise demand for both tradables and non-tradables. Faced with an elastic supply of foreign-produced tradables and an inelastic supply of domestic non-tradables, the relative price of non-tradables would have to increase. Similar to the effect of higher export prices, this effect can also be illustrated by looking at the current account. Higher income receipts from higher net foreign assets imply a higher current account balance which would need to be offset through a lower trade balance, requiring a more appreciated real exchange rate.

• Government consumption or expenditure. Higher government consumption would appreciate the real exchange rate if it includes a higher share of non-tradable goods than private consumption or investment. In this case, the relative demand for non-tradables would rise as private consumption and investment are crowded out by an increase in government consumption and the latter is more heavily directed towards non-tradable goods. The ensuing change in relative demand for non-tradables would change the relative price and thus the real exchange rate.

• Trade liberalization. Trade restrictions raise the domestic price of tradable goods, and thus the overall price level and the real exchange rate. Conversely, a more open trade regime should be associated with a more depreciated real exchange rate.

C. Empirical Framework³⁵

6. This study uses a measure of the real effective exchange rate that is based on the consumer price index and weighted by total trade. The CPI-based real exchange rate is the most commonly available and employed measure in the literature, although other measures, such as those based on the producer price index or unit labor costs, have certain merits and are also frequently used. In particular, the unit labor cost-based real exchange rate is a good indicator to assess the competitiveness of the export sector. However, good measures of unit labor costs in Colombia are not available over a sufficiently long time period. To calculate the effective real exchange rate, this study uses a moving average of Colombia’s total trade—exports plus imports—with main trading partners—most notably the U.S. with a trade share between one and two thirds. Figure 2 shows the behavior of the monthly series of various available real exchange rate measures. Using alternative trade weights can lead to small or moderate differences in the real exchange rate index, but such deviations have remained temporary in the past. The PPI-based index shows a somewhat stronger average depreciation over the past 20 years relative to the CPI based index. However, such deviations would be reversed in the long run, provided that the CPI and the PPI do not diverge systematically.

Alternative Real Exchange Rate Measures, 1990/1–2006/6

(Index: 2000=100; CPI based unless otherwise indicated)

Citation: IMF Staff Country Reports 2006, 401; 10.5089/9781451808919.002.A004

• Download Figure
• Download figure as PowerPoint slide

View Full Size

Alternative Real Exchange Rate Measures, 1990/1–2006/6

(Index: 2000=100; CPI based unless otherwise indicated)

Citation: IMF Staff Country Reports 2006, 401; 10.5089/9781451808919.002.A004

• Download Figure
• Download figure as PowerPoint slide

Alternative Real Exchange Rate Measures, 1990/1–2006/6

(Index: 2000=100; CPI based unless otherwise indicated)

Citation: IMF Staff Country Reports 2006, 401; 10.5089/9781451808919.002.A004

• Download Figure
• Download figure as PowerPoint slide

7. With a view to better detect the long-run relation between the real exchange rate and its fundamental determinants, this study uses a long-annual sample (1962 to 2005). While some variables, in particular exchange rates and the CPI, are available at higher frequencies, other variables are not, or only over a shorter sample period. There is thus a trade-off between sample frequency and length of the sample. Since most relevant variables are rather persistent and the real exchange rate may deviate from its so-called equilibrium for extended periods, it is important to use a long sample in order to be able to detect the underlying long-run relations.

8. As a proxy to capture the Balassa-Samuelson effect, this study uses an index of Colombia’s per capita GDP relative to that of its trading partners. The motivation is mainly the lack of available detailed productivity data rather than strong theoretical underpinnings. However, this should be an acceptable measure as long as relative per capita GDP growth mirrors relative productivity growth of tradables and non-tradables in the long run. This seems to be a reasonable approximation. Productivity growth tends to be higher in the tradable sectors compared to non-tradable sectors, as the former has a larger share of manufacturing while the latter contains a higher share of services. It thus seems reasonable to assume that countries that enjoyed higher productivity growth—and therefore higher per capita GDP growth—had a higher share of that productivity growth in the tradable sector.

9. An export commodity price index is constructed based on price movements of the seven most important export commodities since 1962. The most important one used to be coffee with a total export share of over 70 percent in 1962, which has since fallen to only about 7 percent. The share of oil has fluctuated historically but has now become the most important export commodity with a total export share of about 25 percent. The share of other commodities is significantly less important (coal 12 percent, nickel 3½percent, and gold 2½percent). The index is constructed using price changes weighted by export shares and deflated by the U.S. wholesale price index. Since commodity imports play only a small role in Colombia, import price changes are not considered. An alternative to using the commodity price index would be to use the terms of trade. However, both series are highly correlated and yield very similar results.

10. The variable for net foreign assets used here is a broad measure including all external assets and liabilities, measured in U.S. dollars.³⁶ Net foreign assets are then scaled by exports of goods and services, also measured in U.S. dollars. Another option would have been to scale by nominal GDP. However, nominal GDP depends endogenously on the real exchange rate as the nominal value of non-tradable GDP varies with the real exchange rate, and would thus lead to biased estimates. While exports measured in U.S. dollars could also be correlated with the real exchange rate, the correlation should be smaller than with nominal GDP measured in U.S. dollars.³⁷

11. Government consumption is expressed as a share of GDP. Since both are measured in national currency, correlation between GDP and the real exchange rate should not be a problem.

12. To measure trade restrictions this study uses the implied weighted average tariff rate, calculated as the ratio of fiscal revenue from import duties and licenses over total imports. Good measures of trade restrictions are difficult to come by, especially if long time series are needed. One commonly used alternative measure of trade openness is total trade relative to GDP. However, this measure also suffers from the endogenous correlation between the real exchange rate and GDP. Actual tariff rates could also be used, but unfortunately, data is not available for all years and no data is available prior to 1974.

13. Figure 3 shows the behavior of the key variables described above since 1960. Key features are: (i) a cumulative depreciation of the real exchange rate by more than 50 percent between 1960 and 2005; (ii) a significant decline in relative per capita GDP that took place in two steps, first the 1960s, and then the late 1990s and early 2000s; (iii) a large spike in export commodity prices in the late 1970s due to oil prices and, to a lesser extent, coffee prices; (iv) large swings in net foreign assets, but no pronounced trend; (v) a dramatic increase in public consumption during the 1990s, in part related to constitutional reforms; and (vi) declining trend in tariffs, interrupted by a sharp increase in the mid 1980s followed by a significant trade liberalization in the early 1990s.

Behavior of Key Variables, 1960–2005

Citation: IMF Staff Country Reports 2006, 401; 10.5089/9781451808919.002.A004

• Download Figure
• Download figure as PowerPoint slide

View Full Size

Behavior of Key Variables, 1960–2005

Citation: IMF Staff Country Reports 2006, 401; 10.5089/9781451808919.002.A004

• Download Figure
• Download figure as PowerPoint slide

Behavior of Key Variables, 1960–2005

Citation: IMF Staff Country Reports 2006, 401; 10.5089/9781451808919.002.A004

• Download Figure
• Download figure as PowerPoint slide

14. All variables are fairly persistent and, with the exception of the ratio of net foreign assets to trade, appear to have unit roots. Table 1 shows the results from the augmented Dickey-Fuller unit root test using a specification with trend and intercept. The null hypothesis of a unit root can only be rejected at the 5 (or 10) percent significance level for net foreign assets. Choosing a specification without liner trend does not change the results qualitatively.

Table 1.

Estimation Results with Basic Variables

Note: t-statistics in parenthesis.

Table 1.

Estimation Results with Basic Variables

Specification number	(1)	(2)
Coefficients of the cointegrating relation (long term relation)
Real exchange rate	1	1
Relative GDP index	-0.304	-0.048
	[-0.91]	[-0.16]
Commodity price index	-0.431	-0.296
	[-8.49]	[-5.12]
Net foreign assets to exports	0.010	-0.006
	[0.45]	[-0.27]
Government consumption to GDP	-2.207	-1.775
	[-4.75]	[-2.86]
Average tariff rate		0.243
		[0.48]
Time trend	0.029	0.029
	[14.15]	[ 14.69]
Constant	-1.035	-1.172
Adjustment coefficients in short term equations (speed of adjustment)
Real exchange rate	-0.509	-0.506
	[-3.45]	[-3.38]
Relative GDP index	-0.071	-0.088
	[-1.91]	[-2.27]
Net foreign assets to exports	-1.263	-1.231
	[-2.28]	[-2.25]
Average tariff rate		0.092
		[ 3.08]
Likelihood ratio test for weak exogeneity of commodity prices and government consumption
P-value of LR test	0.68	0.75

Note: t-statistics in parenthesis.

View Table

Table 1.

Estimation Results with Basic Variables

Specification number	(1)	(2)
Coefficients of the cointegrating relation (long term relation)
Real exchange rate	1	1
Relative GDP index	-0.304	-0.048
	[-0.91]	[-0.16]
Commodity price index	-0.431	-0.296
	[-8.49]	[-5.12]
Net foreign assets to exports	0.010	-0.006
	[0.45]	[-0.27]
Government consumption to GDP	-2.207	-1.775
	[-4.75]	[-2.86]
Average tariff rate		0.243
		[0.48]
Time trend	0.029	0.029
	[14.15]	[ 14.69]
Constant	-1.035	-1.172
Adjustment coefficients in short term equations (speed of adjustment)
Real exchange rate	-0.509	-0.506
	[-3.45]	[-3.38]
Relative GDP index	-0.071	-0.088
	[-1.91]	[-2.27]
Net foreign assets to exports	-1.263	-1.231
	[-2.28]	[-2.25]
Average tariff rate		0.092
		[ 3.08]
Likelihood ratio test for weak exogeneity of commodity prices and government consumption
P-value of LR test	0.68	0.75

Note: t-statistics in parenthesis.

15. The long-run cointegrating relation between the real exchange rate and the fundamentals is estimated using the Johansen (1995) maximum likelihood estimator. The estimation is based on the following vector error-correction model (VECM) specification: ${\mathrm{\Delta }y}_t=\eta +{\mathrm{\Pi }y}_{t-1}+\mathrm{\sum }_{i=0}^{p-1}{\mathrm{\Gamma }}_i{\mathrm{\Delta }y}_{t+i-1}+{\epsilon }_t$ where η is an (nx1) vector of constants; n is the number of endogenous variables; p is the number of lags; ε is an (nx1) vector of mean zero normal errors; and Π and Γ_i are (nxn) matrices of parameters. The existence and number of cointegrating vectors depends on the properties of the matrix Γ. If Γ has reduced rank r<n, then it can be written as Γ=AB’, where A and B are (nxr) matrices. The number of independent cointegrating vectors is given by the rank r, and the coefficients of B and A express, respectively, the cointegrating relation and the speed of adjustment.

16. For the set of variables described above, standard tests suggest the existence of one cointegrating vector. If the tariff variable is excluded, both the maximum eigenvalue test and the trace test suggest the existence of exactly one cointegrating vector at a 5 percent significance level.³⁸ If the tariff variable is included, the tests are inconclusive in the sense that the maximum eigenvalue test still suggests one cointegrating vector while the trace test suggests three cointegrating vectors. Based on these results and the fact that more than one cointegrating vector is difficult to interpret economically for this set of variables, the following estimations will be based on the assumption of exactly one cointegrating vector.

D. Estimation Results

17. Estimating the system without the tariff variable generally yields the expected signs of the coefficients, although not all of them are statistically significant. Excluding the tariff variable (only available since 1971) allows the full sample (1962–2005) to be used. The first column in Table 1 shows the main results.³⁹ The coefficients on commodity prices and on government consumption have the expected signs and are statistically significant. A one percent increase in export commodity prices appreciates the real exchange rate by about 0.4 percent in the long run. An increase in government consumption by one percent of GDP implies an appreciation of about 2 percent. Both coefficients are similar in size to estimates obtained in cross-country studies. The effect of relative per capita GDP has the expected sign, but is not statistically significant. Finally, the estimated coefficient of net foreign assets has the wrong sign, although it is not statistically significant.

18. The estimated speed of adjustment of the real exchange rate to deviations from the long run relation is relatively high. About half of the deviation is reversed in only one year. While other studies have found faster adjustment speeds for developing countries than for industrial economies, estimated half lives are usually longer than one year. Consistent with theoretical considerations, both net foreign assets and relative GDP also react to the level of the real exchange rate. If the real exchange rate is above the level implied by the cointegrating relation, net foreign assets fall and GDP grows slightly slower. Both effects are statistically significant. On the other hand, commodity prices and government consumption appear to be unaffected by the real exchange rate level, i.e., they are weakly exogenous.⁴⁰

19. The most striking result of this regression is the high value of the time trend, suggesting structural changes unrelated to the included fundamental variables. The trend indicates an average annual depreciation of 2.9 percent in addition to the estimated effects of the included fundamental variables. The relevant coefficient is highly significant and the model yields poor results if estimated without the time trend. Essentially, Colombia appears to have been experiencing certain structural changes that depreciated the real exchange rate and that are not picked up the fundamentals described above. However, without an economic explanation, it is difficult to judge whether this trend is likely to continue in the future, and, given the size of the trend, the implications for the future real exchange rate are large.

20. Trade liberalization does not appear to be the main reason for the observed trend depreciation. As discussed above, a more liberal trade regime should be associated with a more depreciated exchange rate. Column 2 of Table 1 shows the results of the model including the average tariff rate. In order to be able to use the full sample period, this variable was extrapolated back to 1962, by assuming a constant tariff rate until 1971—the first available observation. The estimated coefficient has the expected sign but is not significant. Moreover, it does not help to reduce the estimated time trend in the model. Using other proxies for the trade regime—such as trade over GDP, or a post liberalization dummy—does not yield better results. One reason for the difficulty in finding a robust effect of trade liberalization could be its correlation with the large increase in government consumption—both took place during the early 1990s (see Figure 3).⁴¹ However, the question remains what other structural changes in Colombia’s economy could explain the trend depreciation.

21. One issue that has played—and continues to play—a central role in Colombia is the security situation. It affects economic performance through various channels. Resources are diverted from productive uses, production costs rise, and uncertainty increases. In addition, it affects government finances, politics in general, and probably capital flows. As the relative GDP index is only an imperfect proxy of relative productivity, measures of security conditions may contain additional relevant information, and thus be empirically related to the real exchange rate. Moreover, security conditions may have an important impact on expected productivity in the future.

22. As a quantitative measure of the security situation, this study uses the homicide rate, for which long time series are available. Figure 4 shows the homicide rate since 1960. Following violent years in the 1950s, the rate gradually fell until about 1970. The emergence of the guerilla and the drug trade lead to rising rates in the 1970s and especially the 1980s. Violence peaked in the early 1990s. Since then it has fallen significantly, interrupted only by a temporary upsurge in 1999–2002, which was followed by a very sharp decline during the last four years.

Homicide Rate, 1960–2005

(Homicides per 100,000)

Citation: IMF Staff Country Reports 2006, 401; 10.5089/9781451808919.002.A004

• Download Figure
• Download figure as PowerPoint slide

View Full Size

Homicide Rate, 1960–2005

(Homicides per 100,000)

Citation: IMF Staff Country Reports 2006, 401; 10.5089/9781451808919.002.A004

• Download Figure
• Download figure as PowerPoint slide

Homicide Rate, 1960–2005

(Homicides per 100,000)

Citation: IMF Staff Country Reports 2006, 401; 10.5089/9781451808919.002.A004

• Download Figure
• Download figure as PowerPoint slide

23. Security conditions appear to a have a strong influence on the real exchange rate in Colombia and can explain a significant part of the trend depreciation. Table 2 presents the estimation results including the homicide rate. If the tariff rate is excluded, all coefficients have the expected signs and are statistically significant. The impact of the homicide rate is fairly large. An increase by 10 per 100,000 depreciates the real exchange rate by 6 percent. It should be noted also that this estimate is robust to excluding other variables. All variables are now statistically significant and the estimated quantitative effects change somewhat when adding the homicide rate. A 10 percent increase in relative per capita GDP implies a 13 percent real appreciation and a 10 percentage point increase in the ratio of NFA to exports raises the real exchange rate by about 0.7 percent. The estimated effects of commodity prices and government consumption are about half of those under specification (1). The time trend now falls to 1.7 percent annually, but stills remains significant. If the tariff rate is also included (specification (4)), the trend drops to only 1.3 percent, less than half of the initial estimate, although the coefficients of commodity prices and government consumption become small and insignificant.

Table 2.

Estimation Results including Security Situation

Note: t-statistics in parenthesis.

Table 2.

Estimation Results including Security Situation

Specification number	(3)	(4)
Coefficients of the cointegrating relation (long term relation)
Real exchange rate	1	1
Relative GDP index	-1.282	-1.253
	[-3.46]	[-4.45]
Commodity price index	-0.145	0.043
	[-2.19]	[0.73]
Net foreign assets to exports	-0.068	-0.080
	[-3.47]	[-4.23]
Government consumption to GDP	-1.213	-0.242
	[-2.12]	[-0.46]
Average tariff rate		0.310
		[0.71]
Security Situation	0.006	0.008
	[ 3.56]	[5.74]
Time trend	0.017	0.013
	[3.79]	[3.72]
Constant	-1.112	-1.224
Adjustment coefficients in short term equations (speed of adjustment)
Real exchange rate	-0.764	-0.759
	[-6.25]	[-5.84]
Relative GDP index	-0.058	-0.076
	[-1.50]	[-1.92]
Net foreign assets to exports	-0.518	-0.877
	[-0.89]	[-1.54]
Average tariff rate		0.127
		[4.55]
Likelihood ratio test for weak exogeneity of commodity prices and government consumption
P-value of LR test	0.29	0.01

Note: t-statistics in parenthesis.

View Table

Table 2.

Estimation Results including Security Situation

Specification number	(3)	(4)
Coefficients of the cointegrating relation (long term relation)
Real exchange rate	1	1
Relative GDP index	-1.282	-1.253
	[-3.46]	[-4.45]
Commodity price index	-0.145	0.043
	[-2.19]	[0.73]
Net foreign assets to exports	-0.068	-0.080
	[-3.47]	[-4.23]
Government consumption to GDP	-1.213	-0.242
	[-2.12]	[-0.46]
Average tariff rate		0.310
		[0.71]
Security Situation	0.006	0.008
	[ 3.56]	[5.74]
Time trend	0.017	0.013
	[3.79]	[3.72]
Constant	-1.112	-1.224
Adjustment coefficients in short term equations (speed of adjustment)
Real exchange rate	-0.764	-0.759
	[-6.25]	[-5.84]
Relative GDP index	-0.058	-0.076
	[-1.50]	[-1.92]
Net foreign assets to exports	-0.518	-0.877
	[-0.89]	[-1.54]
Average tariff rate		0.127
		[4.55]
Likelihood ratio test for weak exogeneity of commodity prices and government consumption
P-value of LR test	0.29	0.01

Note: t-statistics in parenthesis.

24. The estimation results are broadly consistent with other studies on Colombia, although the set of fundamental variables included differs across studies. A recent study by Echevarria and others (2005) finds similar effects for terms of trade, government expenditure, and net foreign assets. They also find a strong time trend of about 2 percent annually and an adjustment speed of about 50 percent annually. The magnitudes of the coefficients are also broadly comparable to Fund estimates derived from a broad panel of advanced and emerging market countries.

E. Implications and Evaluation of the Model

25. The real exchange rate implied by the estimated long run relation tracks the main trends of the observed real exchange rate, but, as expected, the series is much smoother and deviations from the long run relation have often been very persistent. Figure 5 shows the estimated real exchange rate based on the fundamentals, and the interval of plus/minus one standard deviation.⁴² Periods of positive and negative deviations alternate, each lasting about 5 to 10 years and coinciding with the main swings in the real exchange rate. The largest deviations took place in the 1970s and 1980s and average deviations have declined since 1999 when Colombia floated the exchange rate and introduced the inflation-targeting framework.

Actual Real Exchange Rate and Rate Implied by Estimated Long Run Relation

(Index: 2000=100)

Citation: IMF Staff Country Reports 2006, 401; 10.5089/9781451808919.002.A004

Note: Based on model specification (3).

• Download Figure
• Download figure as PowerPoint slide

View Full Size

Actual Real Exchange Rate and Rate Implied by Estimated Long Run Relation

(Index: 2000=100)

Citation: IMF Staff Country Reports 2006, 401; 10.5089/9781451808919.002.A004

Note: Based on model specification (3).

• Download Figure
• Download figure as PowerPoint slide

Actual Real Exchange Rate and Rate Implied by Estimated Long Run Relation

(Index: 2000=100)

Citation: IMF Staff Country Reports 2006, 401; 10.5089/9781451808919.002.A004

Note: Based on model specification (3).

• Download Figure
• Download figure as PowerPoint slide

26. Model-implied real exchange rate paths are similar across specifications, despite differences in the estimated coefficients for the fundamentals. Deviations from the estimated long run relations are provided in Figure 6 for all four specifications presented above. In most years, all specifications yield very similar results and deviations across models are rarely larger than 5-10 percent. However, some individual years show much larger differences, mostly in years with large movements in commodity prices, such as 1976.

Deviation of Real Exchange Rate from Estimated Equilibrium

(In percent)

Citation: IMF Staff Country Reports 2006, 401; 10.5089/9781451808919.002.A004

• Download Figure
• Download figure as PowerPoint slide

View Full Size

Deviation of Real Exchange Rate from Estimated Equilibrium

(In percent)

Citation: IMF Staff Country Reports 2006, 401; 10.5089/9781451808919.002.A004

• Download Figure
• Download figure as PowerPoint slide

Deviation of Real Exchange Rate from Estimated Equilibrium

(In percent)

Citation: IMF Staff Country Reports 2006, 401; 10.5089/9781451808919.002.A004

• Download Figure
• Download figure as PowerPoint slide

27. Deviations from the long run relation have been strongly correlated to current account balances, providing some confidence for the model results. If the estimated model captures actual exchange rate fundamentals well, then large deviations from the fundamentals-implied real exchange rate should be associated with large current account deficits or surpluses. Figure 7 shows the relation for the estimated model. During periods with a more appreciated (depreciated) rate relative to fundamentals the current account tended to show a deficit (surplus). In fact, the relation is quite strong with the estimated model. The correlation coefficient is about -0.7 over the full sample period and even higher than -0.8 since the mid 1970s. Based on this correlation, a 20 percent appreciation not justified by fundamentals would imply a 2½percentage point deterioration in the current account. Since the mid 1970s, all periods with unusually high current account deficits (surpluses) clearly coincided with a real exchange rate that was more appreciated (depreciated) than suggested by fundamentals.

Current Account and Deviation from Long Run Relation

Citation: IMF Staff Country Reports 2006, 401; 10.5089/9781451808919.002.A004

Note: Based on model specification (3).

• Download Figure
• Download figure as PowerPoint slide

View Full Size

Current Account and Deviation from Long Run Relation

Citation: IMF Staff Country Reports 2006, 401; 10.5089/9781451808919.002.A004

Note: Based on model specification (3).

• Download Figure
• Download figure as PowerPoint slide

Current Account and Deviation from Long Run Relation

Citation: IMF Staff Country Reports 2006, 401; 10.5089/9781451808919.002.A004

Note: Based on model specification (3).

• Download Figure
• Download figure as PowerPoint slide

28. Recent movements of the real exchange rate have been driven mainly by improving fundamentals and a reversal of the depreciation in 2002-03. Both commodity prices and the security situation have improved significantly since 2003, while the other fundamentals remained broadly stable. The model suggests that movements in fundamentals justified an appreciation of about 10 percent since 2003 (Figure 8). During the same time, the deviation of the real exchange rate from the long run relation was reduced by about 10 percentage points, as the sharp depreciation of 2003 was gradually reversed.

Equilibrium Real Exchange Rate Movements—Decomposition by Source, 1990–2005

(Cumulative change in percent)

Citation: IMF Staff Country Reports 2006, 401; 10.5089/9781451808919.002.A004

Note: Based on model specification (3).

• Download Figure
• Download figure as PowerPoint slide

View Full Size

Equilibrium Real Exchange Rate Movements—Decomposition by Source, 1990–2005

(Cumulative change in percent)

Citation: IMF Staff Country Reports 2006, 401; 10.5089/9781451808919.002.A004

Note: Based on model specification (3).

• Download Figure
• Download figure as PowerPoint slide

Equilibrium Real Exchange Rate Movements—Decomposition by Source, 1990–2005

(Cumulative change in percent)

Citation: IMF Staff Country Reports 2006, 401; 10.5089/9781451808919.002.A004

Note: Based on model specification (3).

• Download Figure
• Download figure as PowerPoint slide

29. Based on the model estimates, the average real exchange rate was broadly in line with fundamentals in 2005. The best fitting model specification (3) suggests that the real exchange rate was 4–5 percent below the estimated long run relation in 2005, which is clearly within the margin of error (Table 3). Alternative model specifications yield similar results, all within +/- 5 percent of the long-run relation. These estimates are based on current (2005) fundamentals. If medium term projections of the fundamentals are used instead, the range of estimates of the deviation from the long run relation broadens moderately, ranging from about -8 to +10 percent.⁴³ However, under the preferred model specification the estimate remains at about -4 percent. These quantitative results are also broadly comparable to similar estimates based on a panel of advanced and emerging market economies. Exchange rate developments through August 2006 have not significantly changed this assessment as the real exchange rate in August 2006 remained close to the annual average in 2005, despite the large but temporary depreciation earlier in the year.

Table 3.

Deviation of 2005 RER from Estimated Long-Run Relation

(In percent)

Note: Positive value means that RER is more appreciated than est. long-run relation.

Table 3.

Deviation of 2005 RER from Estimated Long-Run Relation

(In percent)

		Model specification
		(1)	(2)	(3)	(4)
Based on 2005 fundamentals		0.3	4.5	-4.6	-3.4
Based on projected 2011 fundamentals
	(higher NFA, lower commodity prices, rest unchanged)	9.6	9.1	-3.8	-8.1

Note: Positive value means that RER is more appreciated than est. long-run relation.

View Table

Table 3.

Deviation of 2005 RER from Estimated Long-Run Relation

(In percent)

		Model specification
		(1)	(2)	(3)	(4)
Based on 2005 fundamentals		0.3	4.5	-4.6	-3.4
Based on projected 2011 fundamentals
	(higher NFA, lower commodity prices, rest unchanged)	9.6	9.1	-3.8	-8.1

Note: Positive value means that RER is more appreciated than est. long-run relation.

F. Alternative Approaches

30. Staff’s current account projections broadly confirm these results. The current account deficit reached 1½ percent of GDP in 2005 and is projected to widen to about 2–2½ percent of GDP over the medium term at the current level of the real exchange rate. Such a current account deficit would be sustainable and would broadly stabilize the net foreign asset position as a share of GDP. Recent trade performance also suggests the absence of competitiveness problems and major imbalances. Growth of non-traditional exports remained at about 15 to 20 percent in U.S. dollars through 2005 and early 2006. While imports have also been growing at similar rates, this has been driven mostly by buoyant investment activity—including foreign direct investment.

31. A cross-country comparison of absolute price levels also yields similar results. Inherent in the estimation methodology applied above is the assumption that the real exchange rate was on average correctly aligned with fundamentals. While this should be a reasonable assumption in a relatively long sample, it nevertheless remains an assumption. One way of avoiding such an assumption is to compare absolute price levels measured at current exchange rates across countries using available cross-country PPP estimates. While comparisons across countries that are at different stages of economic development are not very meaningful—mostly because the Balassa-Samuelson effects appear to be strong and more developed countries generally have significantly higher prices of non-tradables—there is some merit in comparing countries with similar per capita GDP. A comparatively high price level, at current exchange rates, indicates that the real exchange rate is more appreciated than in countries with a similar income level and could be an indication that the exchange rate is more appreciated than justified by fundamentals. However, results have to be interpreted carefully since other factors than income levels might justify the level of the price level and the real exchange rate. Figure 9 shows per capita GDP and the price level (both relative to the U.S.) for a broad sample of 46 emerging market countries.⁴⁴ Despite several outliers, there appears to be clear relation between the relative per capita income levels (measured at PPP exchange rates) and relative price levels, although that relation appears to be weaker in Latin America than in other regions. In 2004, Colombia’s relative price level was about 25 percent below the regression line. Considering the 13 percent real appreciation between 2004 and 2005, this suggests that Colombia’s price level still remains about 10 to 15 percent below the level regression line in 2005. While at the lower end of the range observed in most Latin American countries, this is in line with many other emerging markets and indicates neither an excessively high- nor low-price level or real exchange rate.

Cross-Country Price Level and Per Capita GDP, 2004

Citation: IMF Staff Country Reports 2006, 401; 10.5089/9781451808919.002.A004

• Download Figure
• Download figure as PowerPoint slide

View Full Size

Cross-Country Price Level and Per Capita GDP, 2004

Citation: IMF Staff Country Reports 2006, 401; 10.5089/9781451808919.002.A004

• Download Figure
• Download figure as PowerPoint slide

Cross-Country Price Level and Per Capita GDP, 2004

Citation: IMF Staff Country Reports 2006, 401; 10.5089/9781451808919.002.A004

• Download Figure
• Download figure as PowerPoint slide

G. Conclusions

32. Colombia’s real effective exchange rate has been subject to large movements over time. This includes large low frequency swings around a strong depreciating trend. Accordingly, the real exchange rate was much more appreciated in the 1960s than in recent years. In addition to those longer-term movements, the real exchange rate also experienced significant short-term variation. Over the last 10 years, sharp movements were frequently the result of external shocks, often due to financial crisis in other emerging markets.

33. The statistical model estimated in this paper—linking the real exchange rate to a set of long run fundamentals—yields reasonable results, but cannot fully explain the observed trend depreciation. The estimated coefficients of standard fundamental variables all have the expected signs and reasonable magnitudes. In addition, observed deviations from the estimated long run relation are highly correlated with the current account balance. However, the standard variables, such as terms of trade, net foreign assets, relative per capita GDP, and measures of trade restrictions, fail to explain the observed time trend. It appears that structural changes have taken place that were not related to the set of standard fundamentals. Including a measure of the security conditions, (the homicide rate) improves the results and lowers the remaining time trend. Moreover, the variable is highly significant and robust across specifications. Nevertheless, the interpretation is somewhat ambiguous. It could be a proxy for productivity effects not picked up by the relative per capita GDP measure, but it could also be picking up waves of optimism and pessimism that have affected economic performance and the exchange rate.

34. Based on the model estimates, the real exchange rate is currently in line with historical patterns and a significant part of the appreciation since 2003 can be explained by movements in fundamentals. In 2005, the real exchange rate was close to the estimated long run relation with fundamentals and the current real exchange rate remains close to the average of 2005. About half of the appreciation since 2003 can be explained by higher export commodity prices and improved security conditions. The other half reflects a reversal of the sharp depreciation in 2002-03. Alternative methods, such as an assessment of current account sustainability at the prevailing real exchange rate or a cross-country comparison of the relative price level, broadly confirm this assessment.