Learning from models behaving badly

FOREIGN exchange is the world’s largest and most liquid market. Some estimates put the average daily turnover at $4 trillion (£2.5 trillion). Currency is traded over the counter rather than on an exchange, meaning that the machines do not have to be turned on to get a counter-party to your trade. It is traded 24 hours a day, five days a week – from 20:15 on Sunday night until 22:00 on Friday. Though I wouldn’t recommend it, you can buy euros while Brussels is asleep, and sell dollars before New York wakes up.

With the size, liquidity and ease of access of the global foreign exchange market, it can be easy to just dive in, trade and refine your strategies as you go along. But it can be useful to take a step back from the trading screen and take a bottom-up view (not like that) of the fundamentals that drive currency flows. At the same time, where the real world deviates from the theory can teach you when to tread carefully in the markets.

At its most basic, foreign exchange is dictated by supply and demand. In an efficient market, exchange rates will move to ensure that total demand for a currency equals the total supply. Whenever a domestic currency is sold to buy a foreign currency, demand for that foreign currency and a supply of the domestic currency is created. If a country’s balance of payments moves into a deficit – domestic supply exceeding foreign demand – exchange rates will fall, reducing the price of the domestic currency until the balance of payments is restored to equilibrium.

When economies under the Bretton Woods system of fixed currencies approached the transition to free-floating currencies, their respective monetary authorities had to make attempts to forecast the outcome of this shift, and whether structural models of exchange rates would be of systemic forecasting value. The modern foreign exchange market is a very different place to the early 70s, but work by Kenneth Rogoff and Richard Meese of the Federal Reserve at the time still gives insight into he issues of foreign exchange forecasting. Based on the classic elasticities approach of supply and demand, markets ought to be much less volatile than they are in reality. It is a given now, but the theory is that markets do not just respond to this classic, and easily forecastable supply-demand model, but to market expectations of future developments in the fundamentals – including outputs and money supply.

And this is where foreign exchange rate forecasting starts to become much more complex. In doing so it also throws a spanner or two into the works of foreign exchange forecasting. The first proverbial spanner comes in the form of the link between prices and exchange rates. Pretty much any theoretical model of nominal exchange rates relies on purchasing power parity (PPP). There should be a long-term relation between national price levels and exchange rates. If you can use these national price deviations, you should be able to model foreign exchange rates. The problem is that these PPP rates have a very long- relationship – with a half life of 3 years or more.

The second issue relating to the disconnect between exchange rates and fundamentals was cited by Rogoff himself, as one of the six millennium major puzzles in international macroeconomics. That is, not only do macroeconomic fundamentals fail to explain the effects of exchange rates, but it is not easy to systematically trace back the effects of exchange rates to economic fundamentals either.

So what do these difficulties in forecasting foreign exchange rates teach you, the retail trader? First of all, you should be grateful that it is so difficult. If it was possible to perfectly model and forecast the foreign exchange market, it would be nigh on impossible for the retail trader to make money. Secondly, it highlights the risks inherent in fundamental analysis. Markets are not rational. And, given the same fundamentals of outputs and money supply 12 different times, they will react in a dozen different reactions. That is not to say non-linear foreign exchange modelling cannot produce a workable trading system – the enormous electronic trading market would back this up. But traders trying to model markets should tread carefully.