Neural Networks with Divisia Money: Better Forecasts of Future Inflation

If macroeconomists agree on anything, the most likely candidate would be the long-run relationship between the aggregate price level and the quantity of money or, alternatively, the long-run rates of money growth and inflation. But even this relationship is clouded. Even the relatively close relationships in the 1960s and 1970s, were disturbed occasionally by large shocks to the relative prices of energy and other primary commodities that had disproportionate effects on the price level (see, for example, Tatom, 1981). By the 1980s, however, some as-yet-unknown influence caused the trend rate of velocity to deviate sharply from the nearly constant 3 per cent rate it had exhibited over the previous thirty-five years and undermined virtually all attempts to ‘fix’ the aggregate price equation with further empirical efforts along these traditional lines. As such, the growth rate of the M1 aggregate in the USA (and similar measures abroad) became a very poor indicator of future inflation and led to some large overpredictions of inflation until at least the middle of the decade.