Capstone Gallery — Exemplar Papers Behind the ASSIP Poster Deliverable

This gallery holds five complete, publication-style exemplar papers — full ~3,400–4,800-word research narratives that show what “done” looks like at full size, before the distillation onto a 48”×36” poster and into a 200–300-word JSSR abstract that ASSIP actually grades.

What ASSIP grades vs. what these exemplars show. ASSIP 2026’s terminal deliverable is a research poster + a JSSR abstract + an 8-minute poster talk + a replication packet (see Assessment 8 for the rubric and Lab 8 for the build). A full empirical-finance paper is an optional companion artifact, not a required one — recommended only for students who want a journal-shaped piece for college applications, undergraduate-journal submission, or a working-paper public draft. These five exemplars are written as that optional companion paper at the gold-standard length, precisely because the ASSIP poster is a distillation of the full research narrative, and you cannot distill what you have not first laid out at full size. Read each exemplar as the research story behind a single ASSIP poster: the six panels of the poster (question · why · design · headline · robustness · contribution-and-limits) each map to a section of the exemplar paper, and the JSSR-ready abstract on the poster’s title block is the abstract section of that paper at one-twentieth scale. When you build your own ASSIP poster in Lab 8, you are doing the reverse translation: the full research argument lives in your head (and, optionally, in your companion paper), and you compress it onto six panels and ten abstract sentences.

Each paper runs from abstract to honest limits the way a real empirical-finance paper does: a question worth asking, a literature it sits inside, a dataset, a stated identifying assumption, a headline table, robustness organized around the threats, and a conclusion that says plainly what the design cannot establish. They exist so that before you build your own poster, you can study what the full narrative behind a poster looks like, instead of inferring it from a rubric.

Each paper is paired with an annotated commentary — a “How this paper was built” section at the end — that narrates every choice against the camp’s conventions and the Appendix-D table-and-prose standard: why the standard errors are clustered this way and not that, why the identifying assumption is phrased as one sentence, why a verb was softened from “causes” to “is associated with.” The paper shows you the finished object; the commentary shows you the decisions that produced it. Read both — and then, before you sit down to draft your own poster, do the exercise in your head: which sentence of this paper is the question panel? which is the design panel? which sentence of the abstract is “we find”? That is the translation Lab 8 makes you perform on your own work.

Read this first: every result here is synthetic

All quantitative results in these five papers are synthetic and illustrative. They were constructed for instruction — built to be consistent with the relevant Week-6 notebooks and seeded so they regenerate bit-for-bit — and they are NOT empirical findings. No coefficient, standard error, sample size, correlation, or figure in this gallery describes a real lender, a real firm, a real patent, a real filing, or a real market. Several of the datasets are built with a known, planted truth precisely so the method can be checked against a ground truth that no real dataset gives you — that is the whole point of a synthetic study (Lab 4’s slogan: when you are unsure whether a method does what it claims, build a universe where you know the truth and check).

Read these papers for their craft — the structure, the table design, the calibrated verbs, the disciplined hedging, the honest limits — and copy the craft, never the numbers. Where a real paper would name a data vintage, a journal, or a statute, an exemplar says so and stops. Where a citation could not be verified, it is tagged [CHECK] rather than fabricated, exactly as the conventions require. Every paper carries its own synthetic-data notice at the top; this one governs the whole gallery.

The five capstones

Each paper is tied to a dataset track and to a Gao or anchor paper that defines its question, so you can trace the line from a published study to a camp-scale reproduction.

  1. Fair Lending on HMDA — Track: mortgage-denial decomposition · Anchor: Sun & Gao (2019) · Notebook: nb6.4. How much of the minority/non-minority mortgage-denial gap survives the legitimate underwriting controls? A Blinder–Oaxaca decomposition on a synthetic, HMDA-like panel with a planted amount of discrimination, used to show exactly how an omitted credit score inflates the residual and how over-controlling on a pricing channel collapses it. The model paper for the selection-on-observables discipline: a residual is a bounded, fragile number, not a verdict.

  2. Common Ownership from 13F — Track: institutional-holdings overlap · Anchor: Gao, Han, Kim & Pan (2024), “Overlapping institutional ownership along the supply chain and earnings management of supplier firms,” Journal of Corporate Finance, 84, 102520. When the same institutions hold large stakes in competing firms, does it change how much those firms voluntarily disclose? Builds a firm-pair common-ownership measure from synthetic 13F-like holdings and runs a panel fixed-effects regression of a disclosure index on peer common ownership. The exemplar for disciplined observational work: the association survives the most demanding specification, and the paper refuses to call it causal. Note: this is a student-track variant — the anchor paper’s actual outcome is supplier-firm earnings management (discretionary accruals); the capstone substitutes voluntary disclosure as a sibling outcome that is easier to build at camp scale. See the variant notice at the top of the capstone.

  3. Innovation from USPTO PatentsView — Track: market-based innovation measurement · Anchor: Kogan, Papanikolaou, Seru & Stoffman (2017) · Notebook: nb6.1. What is a single patent worth? Reconstructs the KPSS measure — the abnormal stock return in a three-day window around a patent’s grant, scaled by market cap — on a seeded synthetic panel where the true patent value is known. Recovers the planted value, validates it against next-year profitability, and shows how contaminating news in the grant window degrades the measure. The model for turning an unobservable into an observable and checking the recovery.

  4. SEC 8-K Text Classification — Track: AI / text-as-data module · Notebook: nb6.5 · Out-of-sample validated. Can a classifier recover the type of event an 8-K reports from its text, and does the short-window return reaction differ by type? Treats the classifier as a measurement instrument, not an oracle: hand-labels a gold set, freezes the rubric against a held-out test split before any returns are examined, and reports out-of-sample precision/recall/F1 and Cohen’s κ. The exemplar for using an LLM in research honestly — with the training-data-leakage audit and full prompt disclosure that applied work too often skips.

  5. FRED Macro Event Study — Track: macro-announcement event study · Anchor: Kuttner (2001); Bernanke & Kuttner (2005). Do FOMC surprises move the market? Separates the expected component of each rate decision from its surprise and regresses the announcement-day equity return on the surprise, with HAC (Newey–West) standard errors, placebo dates, and split-sample checks. The model for the full event-study scaffolding from free data — and for reporting an “announcement-day association” under a tightly defended timing assumption rather than overclaiming a causal effect.