Does supply equal demand in every market, all the time? Of course not. But in the long run, it is safe to assume that they do. The type of economic model most appropriate for this perspective is Computable or Applied General Equilibrium (AGE). Obviously the results churned out by this kind of model is not meant to be a predictive "forecast", but rather a representation of the underlying fundamentals determining economic structure and relations. We can call this type of model "smooth". Such models are so smooth, they don't even have money variables - only price ratios matter, not actual quoted prices.
Models like this are not useful for representing important phenomena such as the business cycle. This is because business cycle phenomena are mainly due to rigidities that interfere with the underlying tendency towards equilibrium. Macroeconomic models are filled with lots of relations that imply rigidities - for example, the so-called Phillips equation, which relates inflation with unemployment. These kinds of models are coarse; and they have every appearance of plugging in ad hoc conjectures. (In short, they're just plain ugly.) But the models do predict business cycles and monetary effects; we do get the rough-edge economic realities of the short-run.
The view that smooth models are unable to explain rough realities has been challenged by the "real business cycle" school (whose proponents won the Nobel prize last year). They showed that putting in stochastic elements in a "smooth" model generated time series that appeared to mimic macroeconomic trends.
As a gauge of the influence of this school, coarse models are slowly disappearing from the fashion scene. In come the smooth models, gradually expanding in scope - for example, money (or liquidity) can now be integrated as a time-saving device (Gavin and Kydland, 1998). Increasingly macroeconomic models of the business cycle and monetary aggregates are beginning to look more and more like the smooth, applied general equilibrium models.
Consider the shift by the International Monetary Fund from its Multi-mod model, which is coarse (with a parallel smooth model), to the Global Economic Model, which is semi-smooth (i.e. it's basically smooth, with coarse elements integrated here and there to improve realism). Unfortunately it is not yet a public access model (unlike the Multi-mod) as it remains under development (see the website ).
Was there anything wrong with the old-style models? In terms of short-term predictive ability, I don't see any advantage from the new-style models - hey, if you can average a 1 percent absolute error in one-year forecasts of GDP, you must be on to something! However the crucial advantage with the new-style models is the ability to link business cycle gyrations with micro-behavior that is better understood - behavior such as price adjustment to balance demand and supply, adjustment of quantity bought and sold in response to price, all based on the notions of maximization and equilibrium. A more coherent relationship can be drawn between the short-run and the long-run. In that broader view, perhaps we do have a hope of generating more reliable economic models out of this shift.