Two Kinds of Nothing: What Insignificant Results in Finance Actually Show

Battle Count: 75

The paper is primarily about research methodology and reporting standards in applied finance, not about trading strategies, signal generation, or portfolio construction. However, it has indirect relevance: correctly interpreting null results (e.g., 'factor X has no effect on returns') is important for quantitative researchers evaluating whether a factor is truly dead or merely underpowered. The bounded/vacuous distinction could inform factor research and strategy development decisions. The paper does not propose any trading models, signals, or backtesting frameworks.

The proposed protocol is extremely simple to implement: read the coefficient and standard error from a regression table, compute the 95% confidence interval (β̂ ± 1.96×SE), compare the relevant edge(s) to a pre-stated threshold δ, and classify the null as bounded or vacuous. No programming, data collection, or model fitting is required. The main 'complexity' lies in the judgment call of defining δ, which is context-dependent and requires domain knowledge.

The proposed protocol is highly reproducible as it requires only the reported coefficient and standard error from standard regression tables (readily available in published papers) plus a stated threshold δ. No additional data analysis or proprietary code is needed. However, the paper itself is a conceptual note with a single stylized example rather than a systematic empirical application. The promised companion audit of published null claims in finance journals has not yet been completed.