[mmcky]

Review and revision of the Phillips Curve Tradeoffs suite, checked against The Conquest of American Inflation and the papers it draws on (Cho, Williams and Sargent 2002; Sargent and Williams 2005; Cogley and Sargent 2005; Sargent and Williams 2025/26; Abreu, Pearce and Stacchetti 1990).

Structure

• New lecture phillips_policy_evaluation on Conquest chapter 9: maximum likelihood estimation of the classical and Keynesian adaptive models
• phillips_drifts_volatilities split; the post-2000 update is now phillips_drifts_volatilities_2
• phillips_lost_conquest moved to the end of the suite

Shared across the suite

• One escape story: Sims and Chung escape once and stay; Sargent (1999) and CWS recur; Sargent and Williams (2005) explain the difference by the prior; escapes are frequent at the gains the suite uses
• The E-stability eigenvalue -1/(1+theta^2) made explicit in 98, 102, 103, 105
• Links to the Bounded Rationality suite, ls_learning, kalman, rational_expectations, markov_perf, and the advanced series, with reverse links from ls_learning, kalman, and rational_expectations
• Rewritten notation and gain tables; MarcetSargent1989jet citation key; direct FRED CSV reads; 20 new bibliography entries

Corrections that change results

• Discounted Riccati cross term in phillips_learning and phillips_lost_conquest
• Kalman-filter normalization and the Sims replication in phillips_priors
• Both incentive constraints for Abreu's stick and carrot, the worst-value feasibility check, and the APS starting-set condition
• Better-than-Nash outcomes of optimal misspecified beliefs shown to vanish as the perceived root goes to one
• Vector field of phillips_escaping_nash (the moment matrix lags)
• Escape-route persistence prediction, an exact longer MCMC chain, and Monte Carlo error in phillips_drifts_volatilities
• Real-time data and the January 2026 propositions in phillips_lost_conquest
• Data statements, figures, and the October 2025 gap in phillips_two_stories

Exercises: about fifty, up from 21, many linking lectures

[Copilot]

Note

Copilot was unable to complete this review at the selected tier due to model policy or plan restrictions.

🟡 Changes recommended

The change still has unresolved findings, notably a numerically fragile np.log(np.linalg.det(V)) usage for covariance matrices and a small wording/line-break issue in the prose, so it needs updates before approval.

2 open issues

• Use slogdet for covariance log determinant · New
• Add "by" to the fiftyfold sentence for clarity · New

What changed in this PR

This pull request revises and extends the Phillips Curve Tradeoffs lecture suite, adds a new econometric policy-evaluation lecture grounded in the cited inflation-learning literature, and updates cross-links, ordering, and bibliography support across adjacent learning / bounded rationality material.

Changes:

• Adds phillips_policy_evaluation (maximum-likelihood estimation and policy-evaluation reconstruction; classical vs Keynesian identification) and integrates it into the suite structure.
• Expands and clarifies the self-confirming, misspecification, credibility, and adaptive-learning lectures (definitions, E-stability details, derivations, figures, and a substantial increase in exercises).
• Updates navigation and references (TOC restructuring, additional cross-links/anchors, and new bibliography entries).

Outstanding review items to address before approval:

• lectures/phillips_policy_evaluation.md: make FRED CSV parsing more robust (avoid brittle series_id column indexing; handle “.” missing values consistently).
• lectures/phillips_policy_evaluation.md: replace np.log(np.linalg.det(V)) with np.linalg.slogdet (and guard against non-positive determinants) for numerical stability.
• Minor wording/typos in lectures/phillips_credibility.md and lectures/prospects_bounded_rationality.md.

File	Summary
lectures/​rational_expectations.md	Adds a cross-link from fixed-point rational-expectations discussion to the Phillips self-confirming lecture.
lectures/​prospects_bounded_rationality.md	Adds an anchor for cross-referencing and refines phrasing around empirical shrinkage results.
lectures/​phillips_self_confirming.md	Major expansion: fixed-point framing, uniqueness/E-stability material, refined misspecification discussion, and additional exercises.
lectures/​phillips_policy_evaluation.md	New lecture implementing maximum-likelihood econometric policy evaluation using US data, with supporting code/figures and exercises.
lectures/​phillips_misspecified.md	Clarifies perceived near-unit-root modeling and refactors related computations/exercises.
lectures/​phillips_credibility.md	Strengthens links to learning/bounded rationality and expands learning-rate discussion and exercises (plus a minor typo fix requested).
lectures/​phillips_adaptive.md	Adds derivations (steady-state inflation, sacrifice ratio), clarifies key discussion points, and expands exercises.
lectures/​olg_adaptive_money.md	Clarifies finite-sample behavior of decreasing-gain learning near the self-confirming equilibrium.
lectures/​ls_learning.md	Adds pointers summarizing how the Phillips suite uses least-squares learning machinery.
lectures/​kalman.md	Adds cross-links to Phillips lectures that use Kalman filtering / drifting-coefficient models.
lectures/​_toc.yml	Inserts the new policy evaluation lecture and updates suite ordering (including the drifts/volatilities split).
lectures/​_static/​quant-econ.bib	Adds new bibliography entries and citation keys supporting the expanded lecture suite.

This pull request requires changes.

_{Review effort: Balanced}

[mmcky]

Review summary

I reviewed the whole PR: I re-derived the corrections, executed all 13 lectures, and checked the prose against the outputs. Every item under "Corrections that change results" checks out:

• The discounted Riccati cross term now matches quantecon.LQ to about 1e-13; main's version was wrong.
• The Kalman normalization is correct.
• Both stick-and-carrot incentive constraints are correct.
• The ρ → 1 result is correct.
• The moment-matrix lag in the vector field is correct.
• The E-stability eigenvalue −1/(1+θ²) is correct.

The preview build failed for three reasons. All three are now fixed on the branch, together with some smaller corrections:

Commit	What it does
1f78c91	Fixes the two execution failures. In phillips_lost_conquest, real_time_gap asked ALFRED for a vintage dated today in local time; ALFRED silently drops a date that is still in the future in St. Louis, so CI failed with a KeyError between about 00:00 and 06:00 UTC. In the phillips_drifts_volatilities Exercise 3 solution, NumPy 2 rejects a stacked np.linalg.solve call.
76b7535	Drops the mystnb figure captions from 20 plots inside exercise solutions. These broke the PDF build ("Not in outer par mode"), as on #1101. None of them were cross-referenced.
8a8bced	Seven corrections, listed below the table.
edbf16d	phillips_drifts_volatilities: the archived interest rate is exactly i/(400 − i), not log(1 + i/400), and the text, both plots and the post-2000 splice now match it. Also, the autocorrelation-time helper no longer raises an IndexError on a short chain.
4d8ee89	Copilot's two suggestions (slogdet, and "by a factor of about fifty"); sort=True on the one pd.concat that warns under pandas 3; IID in place of i.i.d.; eight multi-sentence paragraphs split.

The corrections in 8a8bced:

• cpol_ex4: v^R + T → v^R + G.
• cpol_ex5: the solution claimed the optimal punishment becomes "no harsher than Nash reversion" as δ → 0. It actually stays about twice as effective (ratio 1.95 to 1.98), and a cell now prints the ratio.
• sc_ex3: now uses the main text's N = 2**14 grid. The coarser grid printed B(C)/C = 0.21 where the main text prints 0.311.
• phillips_escaping_nash: corrects the attribution of ε = 0.0002.
• phillips_priors: corrects the definition of V*.
• phillips_lost_conquest: removes a leftover remark about the Riccati fix.
• phillips_two_stories: pins the Baxter–King window to 2008–2019, so it cannot drift into the 2021–22 surge as the data update.

Verification. All 13 lectures execute cleanly at the new head in the quantecon environment (NumPy 2.3). The slowest are phillips_drifts_volatilities, at about 7 minutes locally, and phillips_drifts_volatilities_2, at 4 to 6. Apart from the intended changes, outputs are identical to the pre-fix run. phillips_lost_conquest, phillips_two_stories and phillips_policy_evaluation also run under pandas 3.1 without warnings.

A side effect in phillips_drifts_volatilities_2. The splice fix moves its printed posterior summaries a lot; for example, the posterior mean of tr(Q) goes from 0.433 to 0.318. The cause is that the effective sample size of tr(Q) is only about 9. Every statement in the text still holds, because the text quotes only features that are stable across seeds. The tables will also move with each new quarter of FRED data, though. This is item 8 in the follow-up issue.

Follow-up issue. The judgment calls about the economics and framing are collected for Tom in #1106 rather than changed here. They include:

• the flat/steep convention, which affects the Bernanke contrast in phillips_lost_conquest;
• the Pivetta–Reis characterization;
• the discount-factor verdict in phillips_learning;
• seven smaller items.

Left as is. Style-checker flags that main breaks just as often (figsize, lw=, caption length), and the ASCII phi/gamma names in the phillips_lost_conquest and phillips_policy_evaluation code.

🤖 Generated with Claude Code

[mmcky]

Style-guide pass

A follow-up pass against the QuantEcon style guide added four commits on top of the fixes in the summary above:

Commit	What it does
66a32e5	Uses {cite:t} for 89 citations that are part of the sentence. With the author_year style, a bare {cite} renders as a bracket, so the live site reads "[Phelps, 1967] formulated a control problem" and "chapter 7 of [Sargent, 1999]"; these now read "Phelps [1967] formulated" and "chapter 7 of Sargent [1999]". End-of-clause citations, and citations after an author named in the prose, stay as {cite}.
7ca702b	Braces the proper nouns in 11 bibliography titles. The references page sentence-cases article titles, so today it prints "The post-war U.S. phillips curve" and "post-world war II"; Phillips, Great Inflation, Great Moderation, World War II, French Revolution and American now keep their capitals, as do the paper titles quoted inside the Sims, Stock and Taylor comments.
973d488	Drops six embedded plot titles that repeated the figure caption below them. The two_stories titles that carry the book's figure numbers stay, because the prose refers to the plots by those numbers.
371f37a	Renames phi_pi to φ_π and gamma_of_beta to γ_of_β in code, and uses \mathbb{E}, \mathbb{V} and \mathbb{P}, as the rest of the suite already does.

The six lectures whose code changed run cleanly, with outputs unchanged. Both of Copilot's inline comments were addressed in 4d8ee89 and have replies, so they can be resolved.

Left as is: figure size, line width and caption length, which main breaks just as often; and the "Lucas Critique" / "Lucas critique" spelling, which varies across the suite.

🤖 Generated with Claude Code

[github-actions]

📖 Netlify Preview Ready!

Preview URL: https://pr-1105--sunny-cactus-210e3e.netlify.app

Commit: 9862ee8

📚 Changed Lectures

• kalman
• ls_learning
• olg_adaptive_money
• phillips_adaptive
• phillips_credibility
• phillips_credible_policies
• phillips_drifts_volatilities
• phillips_drifts_volatilities_2
• phillips_escaping_nash
• phillips_learning
• phillips_lost_conquest
• phillips_misspecified
• phillips_policy_evaluation
• phillips_priors
• phillips_self_confirming
• phillips_two_stories
• prospects_bounded_rationality
• rational_expectations

---

Build Info

• Workflow: Build Project [using jupyter-book]

[mmcky]

thanks @thomassargent30