MONEY-WAGE DYNAMICS AND LABOR-MARKET EQUILIBRIUM thus raises a cruel dilemma for fiscal and monetary policy. The Phillips curve portrayed the rate of wage change as a continuous and decreasing function of the unemployment rate, with wage increases exceeding typical productivity growth at sufficiently low albeit above-minimum unemploy ment rates. Hence, if prices are tied to marginal or average costs, the smaller the level at which aggregate demand sets the unemployment rate the greater is the continuing rate of infiation Strikingly, Phillips found that the nineteenth-century data pointed to a trade-off between wage increases and unemployment in the same way as contemporary data. Lipsey's sequel(1960) showed a statistically significant Phillips-curve relation for the subperiod 1861-1913. In fact, this early Phillips curve was higher(by about one percentage point) than the Phillips curve he fitted to the period 1929-57. 6 Apparently the cost inflation ten dency, if real, is not""in history; in Britain anyway it may be no worse than it used to be But is the Phillips trade-off real, serious, and not misleading? I shall discuss briefly two challenges to the Phillips curve to which this paper is relevant. The first is the question of whether the slope of the wage increase- unemployment relation is great enough to pose a serious dilemma for ggregate demand policy. Though proponents of an American Phillips curve had tough sledding at first--numerous other variables were held to be important(Bowen, 1960; Bhatia, 1962; Eckstein and Wilson, 1962)- Perry's synthesis (1964)of much of this early work left a quantitatively important role for the unemployment rate(as well as for the profit rate and the rate of change of prices)in explaini nts in U.S. manufacturing. But in 1963 Bowen and Berry(1963)found that the decrease of the unemployment rate was far more important than the level of the unemployment rate in contributing to wage increases. The recent study of annual long-term wage data by Rees and Hamilton (1967)also showed a negligible (and statistically insignificant) relation between the eady-state unemployment rate and the rate of wage increase(though s By contrast, in the pure"demand inflation"of Keynes and the classics, a reduc ion of the price trend could be achieved without cost to output and employment since aggregate demand is necessarily superfluous to begin with "Demand inflatio ay be worth preserving, since a regime of mixed inflation"is conceivable My earlier paper(1961)contains a fairly complete taxonomy of inflations(see also Fellner, 1959). Incidentally, the occasional definition of cost inflation as an autono mous upward shift of the Phillips curve is very awkward and does not imply the policy dilemma"with which inflation analysts were concerned in the fifties 1862-1913 regression(his equation [10]) predicts a 2.58 per cent wage increase an nually, while the 1929-57 regression (his equation [13])predicts a 1.65 per cent 四 the same 3 per cent productivity growth in both periods, for bility would have permitted smaller unemployment in the latter s Table 2(p. 30)is evidence of the early Phillips curves under age increases after World War IIMONEY-WAGE DYNAMICS AND LABOR-MARKET EQUILIBRIUM 681 thus raises a cruel dilemma for fiscal and monetary p01icy.~ The Phillips curve portrayed the rate of wage change as a continuous and decreasing function of the unemployment rate, with wage increases exceeding typical productivity growth at sufficiently low albeit above-minimum unemploy￾ment rates. Hence, if prices are tied to marginal or average costs, the smaller the level at which aggregate demand sets the unemployment rate the greater is the continuing rate of inflation. Strikingly, Phillips found that the nineteenth-century data pointed to a trade-off between wage increases and unemployment in the same way as contemporary data. Lipsey's sequel (1960) showed a statistically significant Phillips-curve relation for the subperiod 1861-1913. In fact, this early Phillips curve was higher (by about one percentage point) than the Phillips curve he fitted to the period 1929-57.6 Apparently the cost inflation ten￾dency, if real, is not "new" in history; in Britain anyway it may be no worse than it used to be. But is the Phillips trade-off real, serious, and not misleading? I shall discuss briefly two challenges to the Phillips curve to which this paper is relevant. The first is the question of whether the slope of the wage increase￾unemployment relation is great enough to pose a serious dilemma for aggregate demand policy. Though proponents of an American Phillips curve had tough sledding at first-numerous other variables were held to be important (Bowen, 1960; Bhatia, 1962; Eckstein and Wilson, 1962)- Perry's synthesis (1964) of much of this early work left a quantitatively important role for the unemployment rate (as well as for the profit rate and the rate of change of prices) in explaining money-wage movements in U.S. manufacturing. But in 1963 Bowen and Berry (1963) found that the decrease of the unemployment rate was far more important than the level of the unemployment rate in contributing to wage increases. The recent study of annual long-term wage data by Rees and Hamilton (1967) also showed a negligible (and statistically insignificant) relation between the steady-state unemployment rate and the rate of wage increase (though By contrast, in the pure "demand inflation" of Keynes and the classics, a reduc￾tion of the price trend could be achieved without cost to output and employment, since aggregate demand is necessarily superfluous to begin with. "Demand inflation" may be worth preserving, since a regime of "mixed inflation" is conceivable. My earlier paper (1961) contains a fairly complete taxonomy of inflations (see also Fellner, 1959). Incidentally, the occasional definition of cost inflation as an autono￾mous upward shift of the Phillips curve is very awkward and does not imply the "policy dilemma" with which inflation analysts were concerned in the fifties. At a constant price level and an unemployment rate of 2 per cent, Lipsey's (1960) 1862-1913 regression (his equation [lo]) predicts a 2.58 per cent wage increase an￾nually, while the 1929-57 regression (his equation [13]) predicts a 1.65 per cent annual increase. At the same 3 per cent productivity growth in both periods, for example, price stability would have permitted smaller unemployment in the latter period. But Lipsey's Table 2 (p. 30) is evidence of the early Phillips curve's under￾estimation of the wage increases after World War 11