Uruguay: Selected Issues Paper

This paper estimates cyclically adjusted balances for Uruguay, and discusses methodological and practical implementation issues. In line with standard practice, this paper assumes aggregate fiscal revenue elasticity equal to one. The study also focuses on the cyclically adjusted primary balance, so interest payments are excluded from the analysis. It also estimates Cyclically Adjusted Balances (CABs) for both the consolidated public sector and the general government. The economic development and the credibility of the inflation target are discussed. This study identifies the drivers of the low profitability of Uruguayan banks.

Abstract

This paper estimates cyclically adjusted balances for Uruguay, and discusses methodological and practical implementation issues. In line with standard practice, this paper assumes aggregate fiscal revenue elasticity equal to one. The study also focuses on the cyclically adjusted primary balance, so interest payments are excluded from the analysis. It also estimates Cyclically Adjusted Balances (CABs) for both the consolidated public sector and the general government. The economic development and the credibility of the inflation target are discussed. This study identifies the drivers of the low profitability of Uruguayan banks.

III. What explains the low profitability of the Uruguayan banking system?1

A. Introduction

1. While Uruguay achieved high economic growth in recent years and managed to avoid a recession during the global financial crisis, its banks have seen a noteworthy trend decline in profitability. Between 2006 and 2009 real GDP growth and credit growth expressed in constant exchange rate terms averaged 6 and 20 percent per year, respectively, but banks’ pre-tax return on assets (RoA) fell from 1.7 percent to 0.5 percent, with more than half of the banks recording losses at year-end 2009. Even in the best profit year of 2007, the RoA remained below the 2 percent mark. At first glance, this result appears surprising because banks have maintained solid credit spreads and recorded low non-performing loans. However, a number of factors have weighted on banks’ bottom line, notably high levels of liquidity and low returns, strongly rising personnel costs, and unfavorable exchange rate developments.

2. This paper identifies the drivers of the low profitability of Uruguayan banks. Using ratio analyses and standard efficiency measurement techniques, it sets out to identify the worsening elements on both the income and cost side, and it derives changes in total factor productivity as well as in efficiency scores. This paper does not yet compare the efficiency of Uruguayan banks to those in other countries with a similar level of banking sector development; however, such a cross-country study for the region is proposed.

3. The results can be summarized as follows. Among the production factors of intermediation—labor, fixed assets and financing—the pronounced rise in real wages has had the strongest impact on profitability. Other factors comprised low returns on the high stock of liquid assets, and to a lesser extent, the appreciation of the Uruguayan peso weighing on the value of the high share of dollar-denominated assets on banks’ balance sheets.

B. Ratio Analysis

4. Bank performance in Uruguay has been influenced by both cyclical and structural developments. Between 2005 and 2009 the pre-tax return on average assets fell from 1.4 percent to 0.5 percent, having peaked at 1.9 percent in 2007. During this time the share of net earnings from lending and provision of services (net financial margin) in total assets dropped by 1½ percentage points, whereas the share of salary costs grew by half a percentage point (Table 1).1 To date, the RoA is not on a path of recovery from its depressed 2009 level.

Table 1.

Revenue and Expense Items

(In percent of average assets)

article image
Sources: Banco Central del Uruguayand author’s calculations.

5. There is widening gap between reasonably profitable and unprofitable banks. In 2009, the standard deviation of banks’ return on assets was one third higher than the 2005-07 average. Loss-making banks had on average a lower financial margin as well as higher labor and administrative costs than profitable banks (a few banks incurred the cost of severance packages). The unprofitable institutions also tended to be smaller, accounting for only one third of total assets, and more dollarized, which required a larger exchange rate adjustment.

6. Overall, the impact of macroeconomic factors on profitability has increased lately. The significant deterioration in the system’s RoA in 2009 is mainly owed to required valuation adjustments for the effects of elevated inflation and the depreciation of the large share of U.S. dollar assets. Totaling 1½ percent of total assets, these additional expenses were charged to banks’ income statement to preserve the real value of equity capital. The increase in valuation adjustments explains two thirds of the deterioration in the RoA from the peak year of 2007, with the reduced financial margin accounting for the balance.

7. Among the cost components, the rising share of salaries stands out. Bank employee compensation rose by 10 percentage points to just under 40 percent of the net financial margin (Figure 1). Given this trend, a number of banks decided to shed labor in 2009 but this was reportedly achieved only through severance packages of as much as four years worth of salaries. Payroll costs accounted for the largest single change in main cost elements apart from the valuation adjustments for inflation and exchange rate variations.

Figure 1.
Figure 1.

Income Statement Items as a Cumulative Share of the Net Financial Margin

Citation: IMF Staff Country Reports 2011, 063; 10.5089/9781455219025.002.A003

Sources: Banco Central del Uruguay, and author’s calculations.

8. The increase in personnel cost was both in total payroll costs and salaries per employee. Starting out from a near zero growth rate in 2005, nominal personnel costs thereafter grew by an average 16 percent for total payroll osts and 11 percent for salaries, which is well above the average inflation rate of 7½ percent (Figure 2) and in the absence of gains in labor productivity (see Section C). The fact that overall costs outpaced average salaries in all years but one implies that employees continued to be added to the workforce. Given the decline in earnings per employee this raises concerns about overstaffing.

Figure 2.
Figure 2.

Growth of Nominal Labor Costs

Citation: IMF Staff Country Reports 2011, 063; 10.5089/9781455219025.002.A003

Sources: Banco Central del Uruguay, and author’s calculations.

9. Real wages and particularly social security contributions have risen considerably since 2007. Salary increases and growth in non-wage labor costs on average exceeded the rise in consumer prices by 3.6 percent and 6.9 percent per year, respectively. In 2009, wages grew by 4½ percent and non-wage costs by 14 percent in real terms, the latter being primarily due to the higher contributions to the sector’s private pension fund and in some cases to severance pay to redundant employees.

10. On the revenue side, high holdings of liquid assets have long prevented a concurrent increase in earnings. The share of liquid assets in total assets2 remained above 50 percent during 2005-09, but falling interest rates and temporarily elevated reserve requirements with low remuneration have depressed the returns on liquid assets in recent years (Figure 3). The persistently high level of liquidity is owed to inertia in the dollarization of deposits, presumably due to risk aversion coupled with long memory effects on the part of depositors. To avoid currency and maturity mismatches, banks are forced place these dollar deposits in liquid foreign currency instruments. Banks reacted to the decline in earnings from dollar-denominated assets and, more recently, to the reduction in reserve requirements by intermediating some of the freed-up peso liquidity and in the process widening the interest rate margin3 by 2½ percentage points compared to end-2007.

Figure 3.
Figure 3.

Interest Rate Spread and Liquidity Ratio

Citation: IMF Staff Country Reports 2011, 063; 10.5089/9781455219025.002.A003

Sources: Banco Central del Uruguay, and author’s calculations.

In sum, rising salary expenses and low interest rates on the high share of liquid assets predominately account for the drop in the return on assets. These two factors alone explain three fourths of the decline in the RoA between 2006 and 2009, with the earnings drop accounting for half of that decline. While administrative costs and other expenses receded somewhat in terms of total assets, the inflation and exchange rate charges in 2009 also weighed heavily on bank profits. Had the inflation charge been imposed evenly during 2007-09, the RoA in 2009 would only have fallen to 1.05 percent rather than to 0.5 percent. Nevertheless, the decline in profitability appears unmistakably linked to both cyclical factors (low returns on liquid assets; elevated rates of inflation) and structural factors (persistently high share of liquid assets; trend increase in personnel expenses both overall and per employee).

C. Efficiency Analysis

11. In the following, two standard procedures of efficiency estimation are applied to corroborate the results of the descriptive analysis. As before, the drivers of bank profitability are explored. However, as the two methods cannot handle negative bank profits as have been recorded lately, instead gross earnings4 are used as dependent variable in the estimations. Independent variables are labor (employee compensation), fixed assets (administrative costs), and financing (total interest expense); see the summary statistics in Table 2. To obtain developments in cost efficiency, these inputs were further broken down into volume (number of employees, number of branch offices, total value of deposits) and prices (salary per employee, administrative cost per branch office, implicit interest rate on deposits).

Table 2.

Values of Output and Input Variables

(Aveage per-bank values, in millions uruguayan of pesos)

article image
Sources: Banco Central del Uruguay and author’s calculations.

Loans to the non-financial sector.

In constant peso-dollar exchange rate.

12. Exchange rate variations have had a significant effect on balance sheet values expressed in local currency, notably the credit stock. As mentioned, exchange rate movements have necessitated sizable valuation adjustments in the income statement. For example, credit growth expressed in peso terms appeared muted during 2005-07 and in 2009 when the local currency appreciated against the U.S. dollar and diminished the value of dollar-denominated loans expressed in pesos. When instead keeping the exchange rate constant in valuing dollar loans (see second numerical column in Table 2), total credit is shown to have grown on average 4 percentage points more between 2005 and 2009. While net interest income is also affected by exchange rate swings, this is less true for other items of the income statement, particularly salaries and administrative costs that are predominately incurred in local currency.

Data Envelopment Analysis and Changes in Total Factor Productivity

13. First, a non-parametric approach to measuring bank efficiency, Data Envelopment Analysis (DEA), is applied. DEA is a linear optimization procedure that uses information on each bank’s input-output mix to construct an efficient production frontier for the banking system as a whole. The efficiency score of an individual bank is then computed as its “distance” from the efficient frontier, which means that it is a relative, not an absolute measure of efficiency (see Annex 1).5 In other words, even the most efficient bank(s) may not be fully efficient if compared to banks in other jurisdictions. The efficiency score can be broken down into technical efficiency and allocative (cost) efficiency, which, respectively, denote the ability of a bank to obtain maximal output from a given set of inputs and the ability to use these inputs in optimal proportions in view of their respective prices. A third score, scale efficiency, measures the part of the technical efficiency score that is associated with a bank’s ability to operate at its optimal firm size.

14. The results of applying DEA to 2009 data illustrate a wide dispersion of efficiency scores across the 13 Uruguayan banks (Table 3). The median score for technical efficiency is the lowest among the three efficiency measures, implying a great distance of more than half the banks to the fully efficient banks (five institutions), under the assumption of variable returns to scale, which permits large banks to have other economies than small banks. This efficiency gap is less pronounced for allocative efficiency and scale efficiency, which means that banks suffer less from a suboptimal input mix with respect to the relative prices of inputs and that most banks in fact operate at a near-optimal size.

Table 3.

DEA Efficiency Results for 2009

(In percent)

article image
Source: Author’s calculations.

15. Appropriately linking such DEA results for several past years permits to calculate changes in total factor productivity (TFP). By constructing so-called Malmquist indices, the year-on-year change in overall bank efficiency can be calculated. Essentially, this output-oriented chain index measures the evolution of TFP of each entity, i.e. the distance between two production points or input-output mixes irrespective of the distance to other banks. In other words, it is the absolute change in efficiency that is computed, not a relative score denoting the distance to the most efficient banks as with DEA.

16. The computed changes in total factor productivity (Table 4) confirm the findings of the ratio analysis. Following considerable improvements in 2005-06, the average change in bank productivity turned out to be -6½ percent during 2007-09 (both in mean and median scores),6 with less unfavorable cost developments in 2008 mitigating the setback.

Table 4.

Changes in Total Factor Productivity of Uruguayan Banks

(In percent)

article image
Source: Author’s calculations.

17. As a consistency check, bank efficiency with respect to managing the credit portfolio was also assessed. Here, as dependent variable was chosen bank credit to the non-financial sector, with dollar loans expressed in constant exchange rate terms. Under this alternative setup banks’ TFP fell by only 1.3 percent during 2006-09, and it actually advanced in the past three years. This leads to the conclusion that internal processes have not been impaired as much as the earnings-based productivity measures suggest. Put differently, it is the lower return from these assets what accounts for the bulk of the decline in efficiency.

Stochastic Frontier Analysis

18. Stochastic Frontier Analysis—the other main tool in efficiency measurement—estimates a specific production function and decomposes the error term into a pure random error and an inefficiency term.7 This approach proposes a solution to the noise measurement problem faced by DEA and other deterministic applications, which is caused by attributing all measurement errors to the efficiency estimates.8 The following equation is estimated:

ln(yi)=xiβ + vi - ui,

where xi is a vector of input variables in logs and the error term consists of two elements: a traditional random error term, vi, and an inefficiency term, ui.9 The technical efficiency of a bank, TE = exp(-ui), while not directly measurable, can be derived using an estimator developed by Battese and Coelli (1988) that predicts the conditional expectation of ui given the value of (vi - ui ). The importance or impact of the inefficiency term is measured by the contribution of its variance to overall variance, γ=σu2/(σv2+σu2), and conveniently expressed in percentage terms. As with DEA, the inefficiency term essentially describes the distance to the firm(s) with best practices on the efficient frontier.

19. Three SFA models are estimated for a panel of 13 banks during 2003-09. As before, the dependent variable is gross revenue, and the independent variables are payroll costs, administrative outlays and interest expenses. In Table 5, the results for model 1 show that the salary and interest variables are highly significant with the expected signs. Salary costs have by far the greatest weight in the regression.

Table 5.

Stochastic Frontier Analysis—Regression Results 1/

article image
Source: Author’s calculations.

Standard errors in parenthesis; ***, **, * denote significance at the 1%, 5% and 10% level, respectively.

20. In addition, models 2 and 3 account for other bank-specific factors that in the previous sections were shown to have affected profitability. The estimation technique permits to include other exogenous variables deemed to have an impact on bank efficiency. In line with the previous analysis, model 2 accounts for banks’ share of liquid assets in total assets, and model 3 includes both this liquidity indicator and the ratio of the exchange rate-related valuation adjustment for assets and liabilities to total gross earnings. These additional explanatory variables turn out to be highly significant with the expected signs and alter the significance of the main variables only slightly (the administrative cost variable does become insignificant).

21. The resulting efficiency scores confirm the findings so far. Table 6 illustrates the drop in efficiency after the top year of 2006. In model 1 the average efficiency score declined by 20 percentage points between 2006 and 2009, reaching a mere 67.5 percent. The deterioration was particularly pronounced in 2009, when the average efficiency score plummeted 13 percentage points. This finding is in line with the results of the ratio analysis and with the calculated sharp decline in total factor productivity.

Table 6.

Stochatic Frontier Analysis1/

(Average efficency scores)

article image
Source: Author’s calculations.

Aritmetic averages. Scores are not weighted for market share of individuals banks.

22. The other exogenous variables—the liquidity ratio and the exchange rate adjustment—have a clear impact on efficiency outcomes. The output obtained from Model 2 shows that controlling for the share of liquid assets increases the average efficiency score by more than 3 percentage points, or put differently, the actual scores are biased downward due to this additional factor. While in the earlier years there was no measurable difference in scores between the two models, starting in 2007 the gap began to widen from 1.4 to 3.3 percentage points, which is an indication of the negative effect of lower-yield excess liquidity on profitability. Adding the exchange rate adjustment variable in model 3 lifts the efficiency score by another 1.2 percentage points. Here again, the gap in scores between models 2 and 3 started widening slightly in 2007. The results corroborate that among the other drivers of efficiency, excess liquidity has had the strongest effect on the bottom line.

D. Conclusions

23. There has been a trend decline in the efficiency of Uruguayan banks. This paper finds that the observed decline in banks’ average return on assets by 1½ percentage points during 2007-09 was associated with a drop in total factor productivity by 6½ percent and in technical efficiency by 20 percentage points.

24. The paper has identified three principal reasons for low bank profitability. Using ratio analysis and two standard efficiency estimation techniques, the paper shows low returns on the high share of liquid assets in banks’ portfolios and rapidly rising personnel costs to account for most of the deterioration in bank earnings since 2005. A third reason for the drop in profits was the accounting loss associated with an adjustment for high inflation in 2009 that in fact corrects for the rise in prices during 2007-09. Valuation effects associated with the exchange rate variations were found to have considerable effects in all years, but on total they cancelled out. This said, whenever the Uruguayan peso appreciates against foreign currency there will be pressure on bank earnings because the still-high dollarization reduces the peso value of the assets and their returns while expenses are mostly paid in local currency and grow well above the rate of inflation.

Low-income banks may be more disposed to take on riskier clients in an attempt to boost profitability, especially in personal and credit card loans as well as in under-banked market segments like small and medium-size enterprises and home mortgage lending. Due to the rising salary costs there is already now a tendency to outsource mass consumer credit to lower-cost non-bank institutions, some of which are less regulated than banks. Although Uruguay has steadily modernized its regulatory and supervisory system limiting exposures and has provided for a large stock of countercyclical loan loss provisions (Adler et al. (2009), Wezel (2010b)), banks could still be affected by a cyclical downturn driving up loan delinquencies and thus curtailing net interest income in those riskier market segments.

25. The potential remedies for improving bank profitability appear clear-cut. First, bank salaries should not continue to rise well above inflation and productivity advances, if any. Investment in technology for automating processes in transaction banking combined with some further cuts in low-skilled labor would help contain labor cost while improving overall productivity. Second, the efforts to induce a significant decline in financial dollarization should be continued so that the necessary valuation adjustments for exchange rate movements, particularly the depreciation of foreign currencies, decrease over time. Banks could more actively promote peso deposits and in the process increase their peso lending. Lastly, low returns on liquid assets, particularly in U.S. dollars, appear to be a cyclical issue but the excess liquidity itself is becoming structural in nature inasmuch as investment opportunities outside interbank lending and investments in government securities are limited. Some opportunities do exist in lending to small and medium-sized enterprises that have no excess to foreign financing and cannot tap the evolving corporate bond market. Notwithstanding some risks, credit to households, particularly mortgage lending, is another promising avenue as wealth creation continues in line with robust economic growth.

Annex 1. An Illustration of Data Envelopment Analysis

The following chart illustrates the concept of technical and allocative efficiency used in Data Envelopment Analysis. A fully-efficient bank that uses two inputs (x1, x2) to produce one output (y) has a certain production possibility frontier denoted by the unit isoquant I. A bank that produces at point a, which is inferior to the optimal production, has a technical inefficiency that is measured by the distance ab or in relative terms, the ratio ab/ao which gives the percentage by which both inputs would have to be reduced by the bank to attain full technical efficiency. Conversely, the technical efficiency score is denoted by TE = 1- ab/ao = bo/ao. Therefore, the efficiency scores are normalized to between 0 and 1, expressing in percentage terms the degree of efficiency with respect to the leading bank(s) or “best practices.”

A03ufig01

Technical and Allocative Efficiency

Citation: IMF Staff Country Reports 2011, 063; 10.5089/9781455219025.002.A003

Allocative efficiency (AE) is depicted by the distance between a point on the isoquant and the isocost line Wwhose slope is the ratio of the input prices, -w1/w2. The additional distance bc represents the reduction in production costs that would be obtained by changing the input mix in favor of using more of the relatively inexpensive factor x2 and, thus, moving along the isoquant to attain the allocatively (and technically) efficient point d.

References

  • Adler, G., M. Mansilla and T. Wezel, 2009, “Modernizing Bank Regulation in Support of Financial Deepening: The Case of Uruguay,” IMF Working Paper No. 09/199 (Washington: International Monetary Fund).

    • Search Google Scholar
    • Export Citation
  • Battese, G.E. and T. Coelli, 1988, “Prediction of Firm-Level Technical Efficiencies With a Generalised Frontier Production Function and Panel Data,” Journal of Econometrics (38), pp. 387399.

    • Search Google Scholar
    • Export Citation
  • Coelli, T., 1996, “A Guide to FRONTIER Version 4.1: A Computer Program for Stochastic Frontier Production and Cost Function Estimation,” CEPA Working Paper, University of New England, Armidale (Australia).

    • Search Google Scholar
    • Export Citation
  • Coelli, T., D.S.P. Rao and G.E. Battese, 1998, “An Introduction to Efficiency and Productivity Analysis” (London: Kluwer Academic Publishers).

    • Search Google Scholar
    • Export Citation
  • Wezel, T., 2010a, “Bank Efficiency Amid Foreign Entry: Evidence from the Central American Region,” IMF Working Paper No. 10/95 (Washington: International Monetary Fund).

    • Search Google Scholar
    • Export Citation
  • Wezel, T. 2010b, “Dynamic Loan Loss Provisioning in Uruguay: Properties, Shock Absorption Capacity and Simulations Using Alternative Formulas,” IMF Working Paper No. 10/133 (Washington: International Monetary Fund).

    • Search Google Scholar
    • Export Citation
1

Prepared by Torsten Wezel (MCM). I thank the very useful comments received from the staff of the Banco Central del Uruguay.

1

All calculations exclude the restructured public mortgage bank Banco Hipotecario del Uruguay.

2

Includes cash, deposits at the Central Bank (including required reserves), and deposits in other banks.

3

The interest rate margin represents a weighted average of peso and dollar rates on loans and on deposits.

4

Gross earnings (“resultado bruto”) are defined as net revenue from financial intermediation (i.e. interest and non-interest revenue less respective expenses from lending and provision of services) less loan loss provisioning and exchange rate-induced changes in the valuation of assets and liabilities.

5

For a more detailed explanation of DEA see Coelli, Rao and Battese (1998). As a non-parametric approach DEA does not correct for measurement errors and other white noise.

6

With 2006 as base year, the drop in productivity (median) is derived as follows: 100*0.825*1.265*0.896=93.5.

7

For a detailed explanation of the methodology see Coelli (1996).

8

While SFA accounts for measurement errors, it requires assumptions about the production function and is subject to issues of econometric misspecifications.

9

While vi picks up the impact of measurement errors and other noise factors on output values, yi, and is therefore iid N(0,σv2), the additional error term, ui, is a non-negative random variable that accounts for technical inefficiency in banks production and is iid truncated at zero of the N(μ,σu2)distribution.

Uruguay: Selected Issues Paper
Author: International Monetary Fund