The evolution of the international monetary system into a regime of flexible exchange rates and the large volatility of these rates during the 1970s have led to a renewed interest in studying the determinants of equilibrium exchange rates. In this thesis, a non-structural approach to modelling the Australian/US dollar using the techniques of VAR econometrics is developed. Absorption, monetary, sticky-price, and portfolio-balance models are used as the basis of the reduced form which is estimated with Australian and US data over the 1970s.
The results of this study suggest the relative importance of asset market variables and price disturbances. In addition to relative price and interest rate disturbances, disturbances in the current account have an important influence on exchange rate volatility. In the unrestricted VAR model, we observe that unanticipated movements in prices, money and the current account will cause a jump in the exchange rate, and then a return to its equilibrium path. The BVAR model, however, implied that unanticipated movements in these variables will, in addition to causing the exchange rate to deviate from trend, steer the exchange rate along a "saddle path" to a new long-run equilibrium. This latter model, therefore, suggests that money, prices, and the current account may be important policy variables. The models track post-Bretton Woods in-sample experience and generates ex post predictions reasonably well. For the within-sample period of March 1972 to March 1986, the VAR forecasts are found to be clearly superior to the naive no-change extrapolated forecasts. However, the position is reverse when the out-of-sample forecasts are examined. These results are consistent with overseas experience that economic fundamentals explain short-term exchange rate volatility poorly when compared to the random walk model. Key Words: Vector Autoregressive model; Bayesian vector autoregressive model; Exchange Rates; Australia; USA.