Formation of Expectations 03-10-2022 1 Imagine a worker who has arrived at the end of some period t - 1, noticed that in the past the price level has moved jerkily up and down without (we assume for the moment) any overall trend or drift. Then he might adopt an error correction method for making adaptive expectations. guess about the future price level would be composed of two parts. One is the actual price level when make your forecast, 03-10-2022 2 and the other is a term that adjusts for your error in the previous forecast. Formally, we can write equation (8) below The first term is the actual price level at t - 1, and the second is an adjustment factor λ times the error you made in forecasting the price level in t - 1. Typically, we imagine that λ lies between 0 and 1. 03-10-2022 3 We now want to see that if expectations are adjusted by an error correction mechanism such as the one given in equation above, the implication is that this period's expected price t-1Pt, depends solely on the history of past prices. This makes it exogenous to this period's actual price determination. It is useful here to perform a Koyck transformation. Changing time subscripts appropriately, we see that equation also describes how expectations were formed about periods t - 1, t - 2, and so on. For example, at the end of t - 2 the worker works out this equation for the expected price for period t - 1: 03-10-2022 4 We can take this last equation, multiply it by lamda and add to earlier equation In symbols 03-10-2022 5 03-10-2022 6 03-10-2022 7 We are now in a position to investigate the consequences of adaptive expectations for the relationship between Pt and t-1 Pt For a start, suppose that the price level had been constant for a long time at Po. Then, suppose that at the beginning of a certain time T (subscript t is a variable, indicating some time period; subscript T is like the proper name of a particular period), the price level jumps up to P1 and stays there indefinitely. At the beginning of T, all the terms on the right-hand side of equation (9) are equal to Po, so the expected price for period T is given by Po, that is, T-1 PT = Po: 03-10-2022 8 Here we have the extreme Keynesian case during the first period after the price shift. Once T is over, however, expectations are formed by equation (9) with t set equal to T + 1. Hence, the first term on the right-hand side for period T + 1 is P1 and not Po: 03-10-2022 9 The size of each step depends on the parameter λ and on the time elapsed since the price level jump. This process continues indefinitely, with the remaining error becoming smaller and smaller In equation (8), the second term on the right diminishes over time to make the difference (Pt - t-l Pt) arbitrarily small. 03-10-2022 10 03-10-2022 11 So adaptive expectations can make sense in an economic environment in which the price level moves up and down in a fairly random fashion, with the possibility of somewhat more permanent shifts in the background. This is the basic assumption about the economic environment in any static equilibrium model of the economy. We analyze the consequences for changes in output and the price level of generally unanticipated exogenous disturbances that can be interpreted as randomly distributed before they actually occur. Adaptive expectations seem appropriate in this environment. 03-10-2022 12 03-10-2022 13 Let us get closer to a definition of rational expectations. We posited an expectations function t-1 Pt = p( Pt). Rational expectations are usually presented the other way around as That is, the realized price level (or whatever variable you are predicting) equals the predicted price level, plus a stochastic error term with mean zero. Your prediction in (19) is said to be unbiased. If the price level is random, that is, a nonzero єt, is a possibility, with a given distribution across possible values 03-10-2022 14 The Law of Iterated Expectations An implication of the unbiasedness of rational expectations is that the expected value of all prediction errors is zero. difference between the realization of a variable and its forecast, say,(Pt, t-1Pt), equals a stochastic term with mean zero. Even though some error is inevitable, it is as likely to be positive as negative.. Therefore, no one can say in advance what sign the error will have. Unbiasedness implies that the expected prediction error is zero for all future periods. 03-10-2022 15 For example, your forecast of the difference (Pt+1 – t-1Pt+1), the difference between the actual value and the prediction two periods earlier, is also a random variable with mean zero and a given variance. Typically, the variance of the errors increases as the forecasting horizon lengthens, but the expected value is always zero. Rational expectations applies not only to variables such as the price level or real output, but also to the predictions themselves. Consider the relationship between the realization of the price level at t + 1 and the forecasts made one and two periods in advance. If the forecasts are unbiased, then 03-10-2022 16 so that Here є 1t+1 and є 2t+1 are two independent "white noise" error terms. If we take expectations through equation as of time t - 1, then the two error terms drop out and the term t-1Pt+1 is unchanged. We are left with 03-10-2022 17 The left-hand expression in above eqn.t-1(tPt+1), is the expectation, made in period t - 1, of the prediction that will be made at time t about t + 1. The right·hand expression, t-1Pt+1 , is the actual prediction in t - 1 of Pt+ 1 So, equation (22) says that the expected forecast equals the current forecast, a relationship known as the law of iterated expectations. 03-10-2022 18 Those who form their expectations rationally conform to the law of iterated expectations because they use all available information and incorporate all "news" as it comes in If forecasts are fully rational, then the prediction error in one period is already taken into account in forming the next period's prediction. Therefore, no one can look at the forecasting errors and find that one error helps to predict the next. An econometrician would say that the errors are serially uncorrelated or that the expected value of the product of two errors is zero. With rational expectations, the expected error is always zero, and the· errors are not linked in any way by serial correlation. 03-10-2022 19 In contrast, under adaptive expectations, we get serial correlation of forecast errors, implying a violation of the law of iterated expectations. The adaptive adjustment formula given in equation (8) is For the rest of this section, a superscript" a" will be used to denote an adaptive expectation. At time T, just after the price increase, the forecast error equals one, as for 'rational expectations, but in period T + 1 the error is λ and at T + 2 the error is λ 03-10-2022 2, and so on: 20 (26c) Correspondingly, the forecasting errors subsequent to the initial surprise at time T are predictably different from zero. For example, if we take expectations through equation (26c) as of time T. we obtain 03-10-2022 21 From this equation it follows immediately that the law of iterated expectations is broken: and forecasts evolve in a predictable way. 03-10-2022 22 The contrast between adaptive and rational expectations is shown in Figure 11-5, which reproduces the paths of the price level and adaptive expectations, with the addition of the path of rational expectations. Adaptive and rational expectations are the same until T, and then they both entail a one-unit prediction error at T corresponding to the genuine price shock. The two paths diverge subsequently as the rational expectation leaps up to match the new price level, and the adaptive expectation adjusts slowly toward it. 03-10-2022 23 03-10-2022 24 The Pros and Cons of Rational Expectations Rational expectations have great appeal to economists, for this hypothesis comes closest to our vision of homo economicus, a person of thoroughgoing rationality in pursuit of his or her maximum expected utility. 03-10-2022 25 If we propose that people and firms are very acute in choosing what to do in anyone period, then they ought to be equally acute in allowing for the future. Each individual has an interest in devoting at least some time and effort to making good predictions, as more foreknowledge cannot leave you worse off It will usually allow you to make better decisions, and even if adaptive expectations, say, lead to optimal actions, then with rational expectations you could simply reproduce that behavior. Or if your rational expectations are more pessimistic than your adaptive ones, given that the rational expectations are unbiased predictions of what will happen, you might at least have a chance to mitigate the bad events to come. 03-10-2022 26