When Dorothy and her friends finally reach Oz, they present themselves to the almighty Wizard, only to eventually discover that he is just an illusion maintained by a charlatan hiding behind a curtain. This seems eerily akin to to the state of Keynesian economics. It does not matter that Keynesianism isn’t working for Obama. It does not matter that it didn’t work for Bush, or for Japan in the 1990s, or for Hoover and Roosevelt in the 1930s. In the ultimate triumph of theory over reality, the Keynesians say all that matters is the macroeconomic model behind the curtain showing that more government spending leads to more jobs and growth. Consider the recent report from the Congressional Budget Office (CBO), which claimed that Obama’s stimulus created at least one million jobs. As Brian Riedl of the Heritage Foundation noted:

CBO’s calculations are not based on actually observing the economy’s recent performance. Rather, they used an economic model that was programmed to assume that stimulus spending automatically creates jobs — thus guaranteeing their result. …The problem here is obvious. Once CBO decided to assume that every dollar of government spending increased GDP…, its conclusion that the stimulus saved jobs was pre-ordained.

But surely this can’t be true, you may be thinking. Our public servants in Washington would not make important policy decisions based on a model that automatically produces a certain result, would they? Peter Suderman of Reason pulls aside the curtain:

[T]hose reports rely on assumption-packed models that effectively predetermine their outcomes; what they say, in essence, is that the stimulus worked because we assume it did. …That’s especially true when estimating government spending’s productive effects, which is accomplished by plugging numbers into a formula that assumes that government spending produces a multiplier—an increased return for every government dollar spent. In other words, it extrapolates from how much money is put in rather than from what has actually come out. And it does so using a formula that dictates that if money is put in, even more money will come out. According to the CBO’s estimates, depending on how the money is spent, one dollar of government spending can produce total economic activity of up to $2.50. What a deal! …[F]or all practical purposes, the same multipliers that were used to predict how many jobs would be created are being used to estimate how many jobs have been created.

Interestingly, CBO’s analysis is completely schizophrenic. Its short-run budget numbers are based on free-lunch Keynesianism that assumes deficit-financed government spending boosts growth, while its long-run numbers are driven by an assumption that government borrowing is terrible for growth (which is why CBO actually claims higher taxes boost economic output — see, for example, Figure 3 of this CBO analysis). It is impossible to know whether the people at CBO actually believe their own work, or whether they are simply trying to please their political paymasters by producing results that (conveniently) match up with political preferences for more spending today and higher taxes tomorrow. You can draw your own conclusions, but keep in mind that CBO is now making the absurd claim that a giant new healthcare entitlement will reduce budget deficits.

But I digress. Let’s now give a defense of the Keynesian model. The folks at CBO and other Keynesians who publish estimates that inevitably turn out to be wrong (Mark Zandi comes to mind) will claim that they are right because they are predicting results compared to what otherwise would have happened. So when they claim that Obama’s so-called stimulus created jobs, they are really saying that the economy would have lost even more jobs if the government didn’t spend all that money. The problem with this approach is that there is no independent benchmark, but this is not why Keynesianism is wrong. Indeed, most of the economic profession relies on this kind of “counterfactual” analysis. Instead, the problem with Keynesianism is that it fails the empirical test. The Keynesians may be good at constructing models, but that doesn’t mean much if the models don’t match the real world. Here’s what Kevin Hassett of the American Enterprise said in recent congressional testimony:

[M]ost economists learned in graduate school that models like those relied upon most heavily by the CBO provide nonsensical results. The reason the original large scale Keynesian Macro forecasting models were discarded by most of the profession is that they make a simple logical error in assuming that individuals do not change their behavior based on the expectation of future policy. …Professor Barro has been one of the primary contributors to the macroeconomic time series literature that has tried to estimate effects from observed economic data, rather than assume affects, as is done by the Keynesian models. …Barro’s analysis is based on econometric evidence, a reliance on experience. The CBO analysis is based almost exclusively on speculation within the context of Keynesian Macro models that were discredited decisively in the 1970s. …Dating at least back to the seminal work of Nelson (1972), economists have known that the empirical time series approach significantly outperforms macroeconomic models in forecasting competitions. …Ashley (1988) compares data-based time series forecasts to those from the large macro forecasters and concludes not only that the time series approach is superior, but that the macro forecasts were so bad that, “most of these forecasts are so inaccurate that simple extrapolation of historical trends is superior for forecasts more than a couple of quarters ahead.” …Finally, one should note that this literature, combined with an earlier public finance literature, raises questions concerning the welfare gain associated with short-term increases in spending. …Browning (1987) finds that the marginal cost ranges widely, between 10% and 300%. Thus, the welfare costs of paying the bill may be greater than the short-term boost to the economy from the most optimistic estimates. This literature would be consistent with Barro’s analysis that suggests the stimulus makes us worse off in the long run.