The Social Costs of Building Reserves

The increase in international reserves is no longer limited to Asia. The reserve holding of most emerging economies have increased sharply. This rise has led to a renewed focus on the costs and benefits of holding large amounts of reserves, and new research that tries to assess the scale of reserve holding needed to limit emerging markets’ exposure to a sudden stop in private capital flows.

Dani Rodrick has argued that current levels of reserve holdings are excessive: “this behavior (of accumulating reserves) is difficult to reconcile with rationality” (p. 3).

Rodrick emphasizes the government reserves accumulation is often the counter-part of private sector flows, and argues that conceptually, the difference between the rate private lenders charge private sector borrowers for short-term external lending and the rate of return on the government’s reserve assets offers the best measure of the social cost of holding reserves. However, given data difficulties, Rodrick uses the historical average of the EMBI spread as a proxy for the private sector borrowing rate in his calculations. Rodrick defines excess reserves as reserve holdings greater than required to cover three months of imports. Using this measure of excess reserves, he presents three sets of calculations, based on spread levels of 300, 500 and 700 basis points. Using 500 basis point as a midpoint he argues that costs of excess reserves in 2004 is almost 1% of emerging economies’ GDP. His ballpark calculations estimate that protection against the sharp output swings associated with sudden stops in capital flows is also worth about 1% of emerging market GDP. However, Rodrik argues that a decrease in the short-term external debt of emerging countries could reduce the risk of a sudden stop at a lower overall social cost.

Like Rodrik, Lipschitz, Messmacheer and Mourmouras try to balance the benefits of reserves against their cost. They find that it is worth paying an insurance premium of around 1% of GDP to reduce the risk of a sudden stop but argue that Rodrik’s analysis overstates the costs of holding reserves, and thus understates the optimal level of reserves for most emerging economies. Their paper reviews three different methods of calculating the costs of holding reserves: 1) the Rodrick’s measure (difference between foreign interest rates at which the private sector borrows and the rate at which reserves are invested); 2) the difference between the marginal product of capital and the return on reserves and 3) the difference between the interest rate on domestic government debt and US treasuries and similar reserves assets (the fiscal cost of holding reserves). They argue that the actual cost of holding reserves is smaller than is usually estimated no matter which measure is used. The first measure - the Rodrick’s measure - overstates the costs, at least for countries with high levels of reserves. Actual spreads on short-term debt for China, Korea, Mexico and Russia were around 100 basis points, well below Rodrik’s estimate of 500 bp.. They reject the second measure on the grounds that reserves can not be used for physical investment in a country. If countries wanted to foster investment the monetary authorities can reduce interest rates or stop intervening in the FX market. Lipschitz, Messmacheer and Mourmouras argue that the difference between domestic and foreign interest rates is the best proxy for the cost of accumulating reserves. They also argue that looking simply at the difference in domestic and foreign interest rates tends to overstate the real cost of holding reserves, as it ignores capital gains (and losses) from exchange rate changes. Since the currencies of most emerging markets have tended to depreciate over time, the costs of holding reserves are often overestimated. For example, the difference between Mexican peso interest rates and US dollar interest rates was 26% between 1978-2005. When the returns are expressed in dollar terms (taking into account exchange rate movements) the differential is only 2% during the same period.

It is worth mentioning their composite reserve indicator. It is based on the idea that reserves ratio should be broad enough to encompass the three key uses of international reserves expressed by the three separate indicators: 1) Reserves/Imports; 2) Reserves/M2 and 3) Reserves/Foreign debt service. In turn, the composite indicator takes into account net reserves as a share the sum of imports ,M2 and foreign debt service. One way to see if the country is holding more reserves than required is to compare the composite indicator with the maximum loss in net reserves. For example, net reserves as a share of the composite indicator for Brazil in 2004 was 26% (increase by 26% of net reserves as a share of the composite as compared to 2003). The maximum loss of international reserves as a share of the composite indicator was in 1986 when Brazil lost almost 30% of its international reserves as a share of the composite. Therefore, the 26% increase would not be enough to cover the maximum loss of 30% suggesting that there is room for increase in reserves. China is the opposite case, where the net reserves as share of the composite indicator is so high that it can cover two times the maximum loss of reserves. The authors explain that the composite indicator does not provide an ideal threshold of optimum level of reserves for all countries. The positive aspect is that it gives a sense of comparison among countries, that is, which ones have enough reserves to cover their maximum loss or not.

Building on work by Garcia and Soto , Jeanne and Ranciere developed a formal model to assess the optimal level of reserves. Like others, their model is keyed off reducing the economic cost of a sudden stop: a sharp falls in consumption and output. Their model implies that the optimal level of reserves is a function of the probability of a sudden stop (П); the output loss in the event of a sudden stop (γ); the debt to GDP ( λ); the return on reserves (r ) the term premium (δ) and the risk aversion (σ). The authors then estimate the values for each one of these parameters such that the benchmark calibration implies an optimal level of reserves of 10% of GDP or 92% of short-term debt, during 1973-2003 period.

The authors then contrast the optimum reserve level with the outcome implied by the Guidotti-Greenspan rule. They reach two main conclusions: 1) Both models are good proxies to how the optimum level of reserves varies with size of sudden stops; 2) Their model capture the fact that the optimum level of reserves is sensitive to the probability of sudden stops, the term premium and the risk aversion parameter. The Guidotti-Greenspan rule does not depend on these parameters and hence cannot capture these effects.

In principle, their model could be used to assess the reserve adequacy of specific emerging market economies. However, they refrain from presenting the results of such a calculation, and instead look at reserve levels in few key regions. As Figure 5 from Jeanne and Ranciere’s paper shows, even in 2003, emerging Asia reserves to GDP ratio was two times the level that emerged from their model, suggesting that much of the post crisis reserve build-up was excessive in light of other changes that reduced their economies vulnerability to a sudden stop. The authors argue that for their model to explain the increase in reserve in emerging Asian countries between 1997 and 2003 the size of either the sudden stop or the output cost would have had to more than double relative to its 1990s observed level. In contrast, their model suggests that Latin America held about the right level of reserves in 2003, and that the build up in Latin reserves from 1992 on was justified.