NOTES FROM
THE DESK.

Saying that CPI moves the market is too broad to test. A useful hypothesis needs an observable input, a defined response, and a time window. For this study, the input is the difference between the reported CPI value and the consensus estimate—not simply whether inflation is high or low.

The first version of the test will separate headline and core surprises, then compare S&P 500 returns across intraday, one-day, five-day, and ten-day windows. I also want to condition the result on the volatility regime because the same surprise may be interpreted differently when uncertainty is already elevated.

The important discipline is to state the rule before examining the outcomes. If I define a “large surprise” only after seeing which threshold produces the cleanest chart, I am describing the sample rather than testing an idea.

An event study can produce different conclusions without changing a single observation. The difference may come entirely from the window used to measure the response. An intraday window isolates the immediate repricing more closely, while a close-to-close return includes information that arrived before and after the release.

Short windows reduce contamination but can miss delayed interpretation. Longer windows capture continuation and reversal, yet they also introduce earnings news, policy comments, and unrelated macro information. Neither choice is automatically correct; each answers a different question.

For the CPI project, I plan to report several windows rather than select one after the fact. The comparison itself is informative: an immediate move followed by a reversal describes a different process from a reaction that continues for five trading days.

A backtest is persuasive only when its logic can be inspected. The return series is the final line of a much longer argument involving data availability, signal timing, portfolio construction, transaction costs, and the decision to include or exclude each observation.

I want every model in this archive to expose those decisions. That means distinguishing in-sample exploration from out-of-sample evaluation, using information only when it would actually have been available, and showing how performance changes after realistic costs.

The goal is not to make a strategy look robust. The goal is to find the conditions under which it stops being robust. A result becomes more useful when its failure modes are documented alongside its strongest period.

Realized volatility describes how much prices moved. Implied volatility reflects the price of protection and the distribution the options market is assigning to future outcomes. They are related, but they are not interchangeable measures.

That distinction matters around scheduled announcements. A large price move after CPI may still be unsurprising if options already priced an unusually wide range of outcomes. The more interesting observation may be the gap between the move implied before the release and the move realized afterward.

I am treating volatility as both a response variable and an information set. That opens a better research question: whether the pre-release volatility regime helps explain the direction, magnitude, or persistence of the equity-market response.