RDP 2015-12: Modelling the Australian Dollar 2. Literature Review

The theoretical and empirical literature on modelling exchange rates is large and varied. In broad terms, the literature attempts to determine the ‘equilibrium’ level of the exchange rate using a set of ‘fundamental determinants’, with this choice of determinants usually being guided by a theoretical framework. However, the determinants, the theoretical frameworks and the concepts of ‘equilibrium’ vary significantly. Moreover, even for the same set of determinants, the mechanisms through which these determinants are expected to affect exchange rates can differ. For example, a given determinant could affect the real exchange rate by influencing the nominal exchange rate, the relative price level, or a combination of the two.

This review will focus on the strand of literature that uses macroeconomic models of exchange rates, as the baseline ECM – which is set out below in Section 3 – fits within this category.^[3] These types of models typically attempt to explain relatively low frequency movements in exchange rates. However, it is first worth noting that there are a range of alternative approaches to exchange rate modelling that are beyond the scope of this paper, but which may, for example, be better-suited to explaining higher-frequency exchange rate movements. A number of these alternative approaches relax the implicit assumption in macroeconomic models that foreign exchange markets are efficient and are comprised of homogenous participants with rational expectations.^[4]

The broad category of macroeconomic models encompasses a number of different approaches. Three of the most prevalent are purchasing power parity (PPP), macroeconomic balance models, and what Clark and MacDonald (1999) term ‘behavioural equilibrium exchange rate’ models.

Perhaps the most basic concept of an ‘equilibrium’ exchange rate is based on the theory of PPP. The PPP concept is a generalisation of the law of one price, which states that, under certain conditions, the price of any particular tradeable good or service should be the same in all countries when expressed in terms of a common currency. As the law of one price should hold for all tradeable goods and services, currency-adjusted price levels in all countries should be the same, and so ‘equilibrium’ real exchange rates should be constant. Given this, empirical examinations of PPP are often carried out by testing whether real exchange rates revert to a constant mean. The results from this literature are mixed, but in general indicate that where PPP is found to hold, real exchange rates revert to their means at best quite slowly (Rogoff 1996; Taylor and Taylor 2004).

One reason for these mixed results is that real exchange rates are typically measured by deflating the nominal exchange rate using a broad measure of relative price levels, such as one based on consumer price indices. This PPP approach is conceptually appealing as it measures the real exchange rate as the price of a broadly representative basket of goods and services in one country relative to another country (or a number of other countries), expressed in a common currency. However, it will include both tradeable and non-tradeable components, and there is no reason to expect PPP to hold for the latter.

In particular, Balassa (1964) and Samuelson (1964) postulated that prices for non-tradeable items – and therefore any real exchange rate that is constructed using a basket that includes those items – should be higher in countries that have relatively high productivity in their tradeable sectors. The intuition is that higher productivity in the tradeable sector will lead to higher wages in the whole economy and hence to higher prices in the non-tradeable sector. Therefore, the overall price level in this economy, and the (broadly measured) real exchange rate, will be higher, relative to that of another economy with lower productivity in its tradeable sector.

As differential trends in productivity can last for extended periods, the Balassa-Samuelson effect can help to explain why real exchange rates do not appear to revert back to a constant mean (or at least only do so very slowly). Nevertheless, both the basic and Balassa-Samuelson augmented notions of PPP are very long term concepts that do little to help explain short- or medium-term movements in exchange rates, particularly in an empirical sense. To this end, large parts of the theoretical and empirical exchange rate literature are focused on identifying short-or medium-term factors that can affect exchange rates.

One such approach is to use a macroeconomic balance (MB) model, which was first popularised by Williamson (1985). These models are also sometimes referred to as fundamental equilibrium exchange rate models.^[5] In these models, the equilibrium real exchange rate is defined as the level that is consistent both with internal and external balance; that is, with output at its potential level and the underlying current account balance at its ‘sustainable’ level. However, as both potential output and a sustainable underlying current account balance are difficult to quantify objectively, the estimation and interpretation of MB models requires a relatively large degree of judgement. For example, some of these models simply make ad hoc assumptions about the sustainable level of the underlying current account balance. Alternatively, in cases where the sustainable current account is modelled more formally, these models still require an assessment of ‘equilibrium’ or ‘desired’ policy settings (Clark and MacDonald 1999; Driver and Westaway 2004).^[6]

Another approach is to use models that attempt to explain the exchange rate based on the observed values of relevant economic variables. These models are sometimes referred to as behavioural equilibrium exchange rate (BEER) models and, consistent with their use of explanatory variables that are measured based on observed rather than sustainable levels, they tend to have a shorter-term focus than MB models (Clark and MacDonald 1999). The types of explanatory variables included in these models vary depending on the underlying theoretical framework used. Examples include: monetary models, which focus on monetary shocks to the nominal exchange rate and so include variables such as nominal interest rates, the money supply or inflation, and GDP or income; and external balance models, which focus on the determinants of the current account balance and so include variables such as the terms of trade, interest rates, the net foreign asset position and the level of government debt or fiscal deficits.^[7] Given the forward-looking nature of foreign exchange markets, such models often incorporate expectations for these variables (e.g. Chen, Rogoff and Rossi 2010).

While BEER models of major floating exchange rates have often been shown to perform reasonably well within sample, they perform less well out of sample. Meese and Rogoff (1983) document this for monetary models, while subsequent papers have tended to confirm this finding for other types of BEER models (Cheung, Chinn and Pascual (2005), amongst others). Nevertheless, one set of currencies that generally offer an exception to the Meese and Rogoff (1983) finding are so-called ‘commodity currencies’, such as the Australian dollar (Gruen and Kortian 1996) and the Canadian dollar (Amano and van Norden 1995). The better out-of-sample fit is likely to reflect the fact that there is a fairly consistent role for commodity prices in explaining movements in these currencies. For example, using time series analysis both Chen and Rogoff (2003) and Cashin, Céspedes and Sahay (2004) find evidence that commodity prices influence the exchange rates of a number of commodity-exporting economies. Cayen et al (2010) reach a similar conclusion using a panel model with a latent factor that is correlated with commodity prices.

Consistent with this, previous RBA papers that have presented models of the Australian dollar have found a significant role for the terms of trade (ToT) – which is driven largely by commodity prices – in explaining the exchange rate (e.g. Gruen and Wilkinson 1991; Blundell-Wignall, Fahrer and Heath 1993; Tarditi 1996; Beechey et al 2000; Stone, Wheatley and Wilkinson 2005).

Footnotes

For a more detailed taxonomy of the different types of macroeconomic exchange rate models, see Driver and Westaway (2004). [3]

For example, the microstructure approach relaxes the assumption of perfect information. It models the exchange rate as a function of the order flow, which is assumed to reflect private information that is subsequently disseminated into the market (e.g. Evans and Lyons 2002). In contrast, the heterogeneous agent approach introduces agents with differing beliefs (e.g. De Grauwe and Grimaldi 2006). Other strands of the literature assume that markets are incomplete, and allow factors such as financial flows and changes in financial intermediaries’ risk-bearing capacity to influence exchange rates (e.g. Gabaix and Maggiori 2015). [4]

Dvornak, Kohler and Menzies (2003) present a MB model of the Australian dollar. [5]

The IMF's External Balance Assessment model is a prominent example of this approach. For information on this model, see IMF (2013). [6]

In the literature, monetary models are often considered to be separate from BEER models. However, we group them together for ease of exposition, given that both attempt to explain fluctuations in the exchange rate using observed values of relevant economic variables. Notable early examples of monetary models include the Frenkel (1976) flexible price model and the Dornbusch (1976) sticky price ‘overshooting’ model. [7]