INVENTORY CYCLE analysis has so far assumed sales forecasts by firms-projecting a constant multiple of either the latest level, or the latest rate of change, of sales into the future. This structure is adequate to show the way in which expectations and the level of business activity interact, but it has a certain ring of implausibility. Other ways of forecasting sales change the behavior of the model, so we may ask if stability is particularly sensitive to the technique assumed, and if more realistic forecasting techniques will enhance or diminish cyclical stability. Here I propose one forecasting technique that pretends to added realism (while remaining sufficiently simple to be incorporated in a formal inventory cycle model) and partly evaluate its effect on cyclical stability. If we are to shift our attention to forecasting models very far removed from naive extrapolation, we must endow the hypothetical seer with information beyond his own past sales levels. A wide choice of alternative assumptions confronts us; casual inspection of forecasting texts will show that some firms find it useful to look at trends in national income or similar aggregates, while others find more specific indicators useful: automobile tire manufacturers use estimates of the stock of cars on the road, for example. But many of these alternatives are specific to a particular industry; here we seek a model that could plausibly be used by most industries, one which will take account of industry differences while using one functional form (for mathematical simplicity). We shall assume that the forecaster knows which sectors of the economy buy the output of his industry, and has some idea of both the trends in

TWO separate policy issues have led to discussion of a preferred rate of inflation: the possible choice between inflation and jobs, and the choice between taxes on money balances and taxes of other kinds. Okun (1971) for the first of these issues and Logue and Willett (1976) for the second, among others, have remarked that the discussion is typically conducted as though a more inflationary policy means a rise from one steady rate to a higher steady rate. The world might not be like that; a rise in the average rate of inflation might mean, inevitably, a rise in its variability. Okun presented some to suggest that countries with higher rates of inflation do experience more variable rates and suggested why it might be so; Gordon (1971) called the evidence into question. Logue and Willett presented further results that supported Okun's position; but their results did not support Okun in the case of highly industrialized countriesthe ones that concerned him. Below I summarize these findings and report additional that supports Okun for highly industrialized as well as other countries. The reason for Okun's concern is that variability of inflation imposes costs on the economy that we should consider when we choose a macroeconomic policy.' Section II supports the claim that there is an empirical relation across countries between the average rate of inflation and the variability of that rate; but it says nothing about the slope, or even the existence, of a functional relation between them that represents an opportunity locus for a single country. In section III, I conclude with a brief discussion of this issue.

We study a standard n-commodity model in which equilibrium positions are characterized by specified inequalities between society's marginal rates of transformation in production and a single consumer's marginal rates of substitution in consumption; these inequalities are exemplified by, but not limited to, excise and subsidies. We explore circumstances under which certain increases in these taxes and subsidies can be said to decrease welfare. In order to do so, we look for conditions under which the equilibrium consumption vector is well defined by a specification of the and subsidies, and find that the conditions required are stringent. Among our conclusions is the proposition that the validity of consumers' surplus measures for analyzing such problems may depend on assumptions that are more strict than their users have realized.