The only honest answer is that we really don't know.
The IMF released its revised estimates for the currency composition of the world's reserves at the end of 2003 and 2004. The most important change? The IMF clearly indicated that it simply has no data on a large (and growing) fraction of the world's reserves. Some central banks, including many key central banks in "emerging Asia" do not report the currency composition of their reserves to the IMF.
The BIS estimate for the currency composition of new reserves added in 2003, an estimate that showed a very large share of dollar purchases (about 90% of the underlying - that is adjusted for valuation gains/ losses - was doing into dollars) looks to be a simple extrapolation of what the central banks who reported information to the IMF were doing. But one of the reporting banks is Japan, and there is no guarantee that the central banks who did not report were doing anything similar.
The same is true of 2004. Though the combination of the US data and the BIS data on the growth of offshore bank accounts suggest a pretty high fraction of the global increase in reserves went into dollars. The BIS recently estimated a $498b (call it $500b) increase in dollar reserves in 2005. If that estimate is right, that works out to about 78% of the overall global increase (leaving out valuation gains). The percent of all new reserves going into dollars in the (less complete) IMF data? Also 78%.
- IMF data. $292 billion reported purchases of new reserves. $263 billion in reported purchases of dollar reserves (90-91% of the total). Roughly $32.5 billion reported purchase of euro reserves. (euro and dollar totals exceed $292 because of rounding and a fall in swiss franc holdings).
- Increase in reserves by central banks that did not report on the currency composition of their reserves (a mix of valuation gains and new purchases) = $238 billion. I would guess new purchases could not be higher than $210 billion, the large euro/ dollar move implies at least $18b in valuation gains.
- BIS estimate* (from their annual report) for the increase in dollar reserves: $441 billion out of a $502 billion total. * this is the data source I generally have used.
- Sum of inflows recorded in the US data ($249) and the increase in offshore bank deposits of central banks ($145b) = $423 billion. See this article in the recent BIS quarterly.
- Estimated valuation adjusted increase in global reserves, all currencies -- $500 billion.
Note than dollar purchases reported to the IMF exceeded the inflows from central banks reported by the US government, even though the IMF data excluded $238 billion of the increase in global reserves.
- IMF data. $349 billion reported purchases of new reserves. $274 billion in reported purchases of dollar reserves (78% of total). Roughly $45.5 billion reported purchase of euro reserves, and $22 billion reported purchase of pound sterling.
- Increase in reserves by central banks that did not report on the currency composition of their reserves (a mix of valuation gains and new purchases) = $322 billion. My $640 billion estimate for total reserve accumulation implies $291 billion of new purchases -- and $31 b of valuation gains. That is just a guess. $31b in valuation gains may be a bit too high.
- No BIS estimate for dollar and non-dollar reserve accumulation in their annual report.
- Sum of US inflows ($395) and the increase in offshore bank deposits ($103) = $498 billion. Data here.
- Estimated valuation adjusted increate in global reserves, all currencies -- $640-650 billion.
Note that the dollar purchases reported to the IMF fell short of dollar purchases in the US data ... the opposite of the 2003 result.
One last note for my friends in the IMF. You all do market analysts a huge favor by reporting your hard work in an obscure appendix, by reporting the data in SDR rather than in dollars (or in both dollars and SDR) and by releasing your global reserve data in basically the same news cycle as your global growth forecasts. You basically bury information that would headline any market report ... that makes it pretty easy for the rest of us to turn data you bury into headlines. We end up getting more credit than we deserve.