- 11 -
1998; Froot and Frankel, 1989; Delong, Shleifer, Summers and Waldmann, 1990]. In
these models, the risk premium and departures from UIP depend on the relative size of
rational speculators vs. other agents in the market. This is the only explanation forwarded
that allows for market inefficiency. The others try to explain how negative coefficients
may arise despite the markets being efficient. The second set of explanations relying on
forecast errors highlight the difficulties in measuring accurately rational expectations
with sample data. Rational systematic forecast errors in sample data may arise because
of: (a) Presence of regime shifts and Bayesian updation of probabilities that regime shift
has actually occurred. (b) Peso problems, i.e. the misalignment of sample moment from
population moments because not all events to which agents accord positive probabilities
have actually occurred in the sample. Lewis (1989) showed that not all the excess returns
could be explained by learning models and Lewis (1995) that the same applies to peso
problems.
A third and in my view the most satisfactory way of explaining the unwholesome
estimates of beta in the unbiasedness equations are non-linearities. These may arise
because of transaction costs [Baldwin, 1990; Dumas, 1992], Central Bank intervention
[MacCallum 1993; Mark and Moh, 2002 and Moh 2002] or because of limits to
speculation [Keynes, 1923; Lyons, 2001 pp 206-220]. Empirical work on non-linearities
includes Leon, Sarno and Valente, 2004; Baillie and Kilic, 2004; Flood and Rose, 1996;
Flood and Taylor, 1996 and Bansal and Dahlquist, 2000. We know that the real world is
not frictionless and that there exist transactions costs, which have reduced dramatically in
the last 10 or 15 years, but still exist. We therefore need to take these into account as
constraints over market induced arbitrage. Baldwin (1990) shows that even small