[gallant_tauchen] Style recommendations left open after the manual review #1089

A review of the new lecture in #1087 against the QuantEcon manual style guide found a handful of items that were judgment calls rather than clear rule breaches. The clear items (bibliography DOIs and title case, figure captions and names, N(0,1), \mathbb{E} and \mathbb{V}, IID, {cite:t}, ^\top) have already been applied on the PR branch. The three below are recorded here as recommendations for a second opinion. @HumphreyYang, could you review and say which, if any, you think are worth doing?

1. Same-series links: {doc} versus markdown links

The manual's linking page prefers [](mle) for lectures in the same series, with {doc} reserved for cross-series intersphinx links. The lecture uses {doc}mle`` and similar in eight places. Every sibling lecture in this repo does the same, and both forms render identically under Jupyter Book, so changing one lecture would make it the odd one out. Recommendation: leave the lecture as is and decide this repo-wide, either by updating the manual to accept {doc} for same-series links or by scheduling a sweep across all lectures.

2. Calligraphic notation: \mathcal{I} and \mathcal{R}

The manual's advice to prefer P over \mathcal{P} when choosing freely is advisory. \mathcal{I}^0 is Gallant and Tauchen's own symbol for the asymptotic variance of the auxiliary score and appears 28 times; a plain I would read as an identity matrix in a lecture full of matrix algebra. \mathcal{R} names the structural parameter space twice, and R is already the replication count in the Exercise 2 code. Recommendation: keep both.

3. Three random number generators in the main text

The manual's code page asks for one rng created once and reused, with local generators only inside self-contained exercise solutions. The main text creates generators with seeds 0, 1 and 5 for the AR(1) data, the fresh-shock criterion curve, and the stochastic volatility data. Consolidating to one generator would change every printed number downstream, and the prose quotes many of them (kurtosis, the GARCH persistence, every estimate and standard error, the Exercise 4 ratios). Recommendation: leave it, since the cost is a full re-verification of the prose for no change in what the reader learns. If a sweep ever reseeds the lecture, the ratios in the Exercise 4 solution are the sentences to re-check first.

Also noted, outside the style guide

Two factual pointers could not be checked against the papers during the review and are worth a glance by someone with the PDFs to hand: the characterization of Duffee and Stanton (2008) as finding wider confidence intervals with worse coverage for score matching, and the references to Sections 4.1 and 4.2 of Gallant and Tauchen (1996) for the GARCH remark and the stochastic volatility model.

Context: see #1087 for the lecture and the review discussion.

🤖 Generated with Claude Code