The accuracy of a simulated currency index or basket should increase when it is modelled on percentage of traded volumes of currency pairs, see here: http://www.bis.org/publ/rpfx13fx.pdf
, pp 5, 10 and 11. Ideally this index would be dynamic and change as the percentage of traded volumes changed, however that depends on how available that information is and there does not to be a live source of such data. Using the example of FXCM's currency baskets on Mirror Trader (http://www.forextrading.com.au/products ... y-baskets/
), rather than distributing equally trading volumes, it would be better to weight the trading volumes. As a more specific example, with the US currency basket at the moment if you invested 1 lot per trade it would invest 1 lot for each currency pair (EURUSD, USDJPY, GBPUSD, AUDUSD). Going by the results in the 2013 Triennial Bank Survey on p 10-11, t seems like it would be more accurate to invest 24.1% in the euro, 18.3% in the gofer, 8.8% in the cable, 6.8% in the Aussie, 3.7% in the Loonie, 3.4% in the USD/CHF and so on. The more currency pairs used, the greater the diversification of the portfolio, the lower the risk and so there should be higher risk-adjusted returns.