ifp_egm_transient_shocks#

  • Series: lecture-dp

  • File: lectures/ifp_egm_transient_shocks.md

  • Audit date: 2026-08-26

  • Corpus snapshot: c30490a2f4

  • Categories audited: writing, math, code, figures, references, links, admonitions (JAX out of scope)

  • Overall score: 7.9 / 10

  • Priority: HIGH

Score breakdown#

Category

Score

One-line note

Writing

3.5/10

qe-writing-006 ×9; qe-writing-003 ×2; qe-writing-002 ×4, +1 more.

Math

9.5/10

qe-math-009 ×1.

Code

7.5/10

qe-code-001 ×4; qe-code-004 ×6.

JAX

out of scope

JAX rules target lecture-jax.

Figures

5.5/10

qe-fig-005 ×7; qe-fig-003 ×3; qe-fig-008 ×10, +1 more.

References

9/10

qe-ref-001 ×1.

Links

10/10

no mechanical violations detected.

Admonitions

10/10

no mechanical violations detected.

Issues#

Critical#

None found.

High severity#

  • [qe-code-004] — Use quantecon Timer context manager. Count: 6. Lines: 590, 592, 595, 598, 601, 604. Example: time.time(.

  • [qe-fig-005] — Descriptive figure names for cross-referencing. Count: 7. Lines: 365, 616, 631, 795, 933, 982, 1048. Example: code-cell figure without mystnb figure metadata.

  • [qe-fig-008] — Use lw=2 for line charts. Count: 10. Lines: 368, 369, 618, 619, 656, 658, 937, 944, 994, 1068. Example: plot() without lw=.

  • [qe-writing-006] — Capitalize lecture titles properly. Count: 9. Lines: 59, 69, 174, 189, 377, 384, 814, 821, 894. Example: H2 Title Case: ‘The Household Problem’ (Household, Problem).

Medium severity#

  • [qe-code-001] (reviewer) — Follow PEP8 unless closer to mathematical notation. Count: 4. Lines: 656, 842, 848, 862. Example: line 656 writes + y_bar(k) , label=label with a space before the comma (E203); line 842 writes jnp.arange(1, n+1) unspaced while line 843 directly below writes (n + 1) / n spaced; line 848 writes p: float=0.01, which needs spaces around = on an annotated parameter (E252), where 514 in the same file gets it right; line 862 leaves trailing whitespace after return wealth_top / wealth_total (W291, also at 309, 473, 638-639).

  • [qe-fig-003] — No matplotlib embedded titles. Count: 3. Lines: 805, 940, 947. Example: .set(xlabel=‘assets’, title=.

  • [qe-writing-002] (reviewer) — Keep writing clear, concise, and valuable. Count: 4. Lines: 34, 104, 616, 816. Example: line 34 reads ‘we continue extend the IFP’ - a verb is missing; line 104 ends ‘where \(a_y, b_y\) are positive constants’ with no full stop and a trailing space; line 816 has ‘Lets’’ for ‘Let’s’ plus four words of padding (‘some standard measures of this phenomenon’). The fourth is duplication: the figure cell at 616-623 is byte-identical to the one at 365-373, labels included, and both plot a_vec/c_vec - the NumPy solution from line 358 - so the JAX section’s only policy figure is a second printing of the NumPy section’s. The prose around it (426-427 repeating 234-235 verbatim, the identical operator docstrings at 268-275 and 436-443, ‘Let’s road test the EGM code’ at both 349 and 544) has the same copy-forward character.

  • [qe-writing-003] (reviewer) — Maintain logical flow. Count: 2. Lines: 556, 968. Example: lines 556-570 present a verification that cannot work as described. create_ifp is defined twice - at 210 with β=0.96 and again at 401 with β=0.94 - so ifp_numpy (352) and ifp (547) are different models, yet 567-568 concludes ‘These numbers confirm that we are computing essentially the same policy’. Line 570 then diagnoses the gap as ‘mainly due to different Monte Carlo integration outcomes over relatively small samples’, which is a specific wrong answer: both constructors draw 100 standard normals from the same seed, and the discount factor is what differs. Second, exercise ifp_egm_ex1 at 968 says to step r through np.linspace(0, 0.016, 4) and its own solution at 984 uses np.linspace(0, 0.04, 4), justified by a comment (‘With β=0.96, we need R*β < 1, so r < 0.0416’) that names a discount factor the model in scope does not have.

  • [qe-writing-008] — Remove excessive whitespace between words. Count: 2. Lines: 570, 706. Example: 2 spaces.

Low severity#

  • [qe-fig-001] — Do not set figure size unless necessary. Count: 1. Lines: 934. Example: figsize=.

  • [qe-math-009] (reviewer) — Choose simplicity in mathematical notation. Count: 1. Lines: 99. Example: \mathsf Z at line 99 for the Markov chain’s state support - the one decorative letter left in this file, and the same object the code calls z_grid. Plain \(Z\) would do, and would also settle the disagreement across the series, where {doc}`ifp_egm` writes \mathsf S/\mathsf Z and {doc}`ifp_advanced` writes \mathbf S and \mathscr C for the same family of objects.

  • [qe-ref-001] — Use correct citation style. Count: 1. Lines: 1017. Example: {cite} in narrative flow: ‘{cite}’.

Strengths#

  • The Monte Carlo step is written into the math before it appears in code: equation cfequmc at 248-256 restates cfequ_ts with the integral replaced by \(\frac{1}{m}\sum_{\ell=1}^m\), and line 258 says each \(\eta_\ell\) is a standard normal draw - so η_draws in the constructor at 418 is implementing something the reader has already seen defined.

  • Every level of the three-deep JAX operator (458-481) carries a one-line docstring naming exactly what it returns, down to the innermost ∫ u'(σ(R s_i + y(z_k, η'), z_k)) φ(η') dη' at 462 - the nesting is followable without re-deriving it, and the same structure is mirrored by the explicit quadruple loop in the NumPy version at 292-315.

  • y_bar is honest about being an approximation: the docstring at 638-644 gives the double sum it computes, and the display at 673-675 writes the same thing in math as \(\bar y(z) := \sum_{z'} \frac{1}{m}\sum_\ell y(z', \eta_\ell)\Pi(z, z')\) rather than as an idealised expectation, with line 677 saying so in words.

  • The lecture reports its own failure and traces it: the wealth distribution is called implausible at 809-811, the Gini and top-share numbers are ‘a long way out’ at 875, calibration is ruled out as the explanation at 882-887 in favour of a structural impossibility result, and the reader is pointed forward to {doc}`ifp_advanced`.

  • The interest-rate experiment at 903-958 states the limit of what it shows - ‘the differences are minor and we cannot increase \(r\) much more without violating the stability constraint’ - instead of stopping at the direction of the effect.

  • Density and CDF conventions are followed and made explicit: line 124 introduces \(\phi\) as ‘the density of the shock \(\eta_t\) (standard normal)’ in lowercase, and every subsequent \(\phi(\eta')d\eta'\) in 116-117, 147-148 and the code comments at 295 and 473 uses it consistently (qe-math-015 (proposed) clean).