I believe most of us would agree the Federal Reserve's stress tests are an imperfect tool. In fact, over a long enough period of time, they are likely to fail. But that does not mean the Fed should stop doing them.
Bank supervision is an art, not a science, and stress testing is another color in the palette. It does not replace existing supervision efforts. In fact, stress tests are done in addition to day-to-day supervision.
Are they baloney? No way.
While it is true the Fed's current tests are designed in a rather generic macroeconomic fashion, each bank is supposed to also come up with its own module that addresses a specific scenario considered particularly damaging for that bank. The Fed has examiners review and approve this scenario.
In general, while the stress tests will never guarantee a bank is risk free, and thus, are not a stand-alone solution to the "too big to fail" problem, they increase a bank's resilience to economic shocks. Because the Fed tests all parts of the bank simultaneously, nothing is completely in focus, but nothing is completely ignored either.
Through this process, the Fed obtains dramatically better data about the banks and the economy. The Fed stands a much better chance of seeing (and measuring) growing risks in real time, something that it only dreamed of before. And the banks know that the Fed has this data and, as such, is able to monitor them much more closely than before.
The standard for passing is more challenging in the U.S. than in Europe. This is because the Fed directly collects the detailed loan level data and uses it to run its own models, providing a truly independent view of risk. The European Union, conversely, relies on the banks' model to tell them where the risks are. It's easy to see how that practice remains ineffective.
Having followed the stress tests since they were initiated in 2009, my personal observation is that the current iteration is the most demanding test the Fed has put banks through, and that it has a real impact on how they run their business. The bottom line is that to be effective supervisors the Fed can't just strong-arm from a distance. It has to actually change behavior. According to my conversations with Fed officials, stress tests have had more direct impact on banker behavior than anything else seen thus far. This is because the tests provide feedback and information that the vast majority of banks would not have had otherwise.
In the past, if banks stress-tested (and very few did), it was done using a single stressor, such as interest rates, or default frequency. The post-2008 stress tests used by the Fed are multi-factorial as are real recessions and financial crises. Knowing how their bank would fare under adverse scenarios allows banks to modify their behavior and avoid becoming part of the problem. The stress tests are more than a test; they are a lesson.
The Fed does use tools like regression, which (yes) relies on historical data to predict the future. The Fed doesn't expect history to repeat itself, but as the old saying goes, the Fed expects it to rhyme.
The point of regression is to observe historical relationships between variables (such as FICO and default, or Expected Default Frequency/Probability of Default and default or any combination of literally hundreds of variables). Its use does not mean the Fed is a slave to the past; it means the Fed can learn from the past. It does not mean the Fed is trying to predict the future; it means the Fed is trying to be ready for it.