You can collect data from a variety of sources. For example, you can get data from:
• Global Financial Data1
• OECD Stats
• WEO Database
When monthly data are not available, use quarterly/yearly data and retrieve monthly
observations by forward filling, i.e., keep the latest observation available until a new one
is made available.
Using at least 30 years of monthly (end-of-month) data on major currency pairs (for ex
ample, AUD, CAD, CHF, DEM-EUR, GBP, JPY, NOK, NZD, and SEK relative to USD),
construct the following monthly rebalanced strategies2
• Carry trade strategy: On each month t, you buy the top 3 high-yielding currencies
and sell the top 3 low-yielding currencies. This is equivalent to sorting on the basis
of the one-month forward premia at time t.
• Mmomentum strategy: On each month t, you buy the top 3 winner currencies and
sell the top 3 loser currencies. This is equivalent to sorting on the past one-month
exchange rate returns, i.e., the exchange rate return between months t and t − 1.
Put differently, you buy (sell) those currencies that have appreciated (depreciate)
the most over the past month.
• Value strategy: On each month t, you buy the top 3 undervalued currencies and sell
the top 3 overvalued currencies. This is equivalent to sorting on the past five-year
exchange rate returns, i.e., the exchange rate return between months t and t − 60.
Put differently, the currencies that have appreciated (depreciated) the most over the
past five years are likely to be overvalued (undervalued), and you want to sell (buy)
1. Can you present summary statistics and plot the cumulative excess returns?
2. How do these strategies behave during periods of high and low volatility?
3. Would an equally-weighted combined strategy outperform any individual strategy?
Take the USD/EUR currency pair, and consider the following specifications
1. RW with drift (i.e., your benchmark model),
Setup. Using a 10-year rolling window, generate out-of-sample (OOS) forecasts using
the following predictive regression:
yt = α + βxt−1 + εt
where yt is the log exchange rate return between months t − 1 and t, and xt−1 is a one
month lagged predictor observed at time t.
• For RW, set xt−1 as in Slide 17 (Lecture 6),
• For TRa, set xt−1 as in Slide 18 (Lecture 6), extract the potential output using a linear
trend, and compute the output gap as ln(real output) minus ln(potential output).