But raw returns are too crude a measure of investment performance, in which case calculating the Sharpe ratio seems more sensible, giving rise to the loss function

L

g t+3

=

1 P

ˆ 3 g + t µ

ˆ ( 3 ∑ g + t µ

− µ ^{g })^{2 }

; g = 0,1

(10)

Finally, we consider the skewness of the carry trade. It has often been remarked (Brunnermeier et al. 2009, Sy and Tabarraei 2009) that carry trade profits are persistently positive subject to occasional crashes: “going up the stairs and coming down in an elevator.” Thus, it seems important to compare investment strategies that, while having similar rates of return, nevertheless produce distributions with different skew. These results in a loss function that can be defined as

# L

g t+3

=

P(P −1) P−2

ˆ ( 3 g + t µ

) 3 − g µ

1 ⎜ ⎝ ⎛ ∑ P

ˆ ( 3 g + t µ

− µ ^{g })^{2 ⎞3/ 2 }⎠

; g = 0,1

(11)

4

# The Dynamics of the Carry Trade

This section investigates the dynamics of carry trade profits with the local projection approach described in Section 3. Specifically, our results are based on the collection of fixed-effect panel regressions described in expression (1), replicated here for convenience,

m_{j,τ +h }

∆ + = , e j h e h j τ β α

+ ( h π π β

* j ,τ

+ − ( ) , , q j h q j τ τ β π

− q) +

K

β ( h i i

* j ,τ

τ τ β , ) h v j V X i + + − ∑

τ γ , k j h k y ∆ −

+

u

_{j,τ }

k =1

and with LP(m_{t+h }, quarters. Ultimately,

x_{τ ) }we

x β are h = more for

x∈{∆ e, (π ^{* }interested in

−π ),q,(i^{* }− i),VX } and h = 1,K,6 the drivers of the forecasting results

14