Table of Contents — Empirical Finance from the Ground Up

A Textbook, Lab Manual, and Problem-Set Bank for the GMU 8-Week Empirical-Finance Research Camp

Host: Prof. Lei Gao, Associate Professor of Finance, Costello College of Business, George Mason University Feeds: ASSIP 2026 (Aspiring Scientists Summer Internship Program) (GMU College of Science) Audience: mathematically gifted U.S. high-school students (AP Calc BC + AP Stats; some Python)

How to read this TOC. Every line item carries a word-count estimate [~N w]. Prose chapters count finished prose only (worked examples included; code listings counted as their explanatory prose). Problem sets count problem statements plus the fully worked solutions that live in Appendix E. Notebooks count their markdown narration + code comments + “Your Turn” prose (cell counts shown separately, e.g. ~22 cells). Per-week subtotals and a grand total appear at the end. Citations follow CONVENTIONS.md §6; anything unverified is tagged [CHECK].

FRONT MATTER

#	Item	Description	Est. words
FM-0	Title page, copyright, dedication	GMU / Costello / College of Science branding; CC-BY-NC license note	300
FM-1	Preface (by Prof. Gao)	Why empirical finance, why high-schoolers can do real research, what the camp feeds into; grounded in verified CV facts	1,700
FM-2	Articulation Matrix (Camp ↔︎ ASSIP 2026)	2-page matrix: 8-week camp vs. the real ASSIP 2026 program (8 wks full-time (Mon-Fri 9-5), Mon-Fri 9-5, Jun 22 - Aug 12, 2026; CRSP/Compustat/EDGAR/Bloomberg; Claude/ChatGPT; Journal of Student-Scientists’ Research (JSSR), GMU Mason Publishing + JSSR (Journal of Student-Scientists’ Research); 3 GMU College of Science credits + $1,299 tuition (full waivers for Pell-eligible / reduced-lunch students)). Shows where the camp deepens vs. where ASSIP 2026 leads	1,600
FM-3	How to Use This Book	Daily rhythm; how chapters/problems/notebooks/labs/mentor sessions interlock; the reveal-the-trick structure; the recurring student cast (Maya, Devon, Priya, Sam, Leah)	1,800
FM-4	Prerequisite Self-Test	20 questions (calculus, probability/stats, light Python) + full solutions + routing table (“missed these → read Appendix A/B first”)	4,500
FM-5	Notation & Conventions quick-card	One-page reproduction of the `CONVENTIONS.md` notation table + spec-discipline checklist	700
FM-6	Roadmap diagram + dependency graph (prose)	Narrative walkthrough of the 8-week arc and how skills compound	900

Front Matter subtotal: ~11,500 words

THE 8-WEEK ARC

Each week contains: (1) a week-opening narrative, (2) five chapters, (3) five daily problem sets (~6 problems each; solutions in Appendix E), (4) a notebook per chapter with a “Your Turn” extension, (5) a lab manual (Weeks 1–4, 7–8) or reading guides (Weeks 5–6), (6) a 60-minute Lei Gao mentor session, and (7) an end-of-week assessment with rubric.

WEEK 1 — Probability, Sampling, and the Logic of Inference

Theme: Rebuild probability and frequentist inference from the ground up, using simulation as the microscope, so OLS in Week 2 rests on solid foundations.

W1 Opening narrative — From “the average return was 8%” to “what could we have seen instead?”: randomness as the object of study. [~1,300 w]

Chapters - Ch 1.1 — Joint, Conditional, and the Two Laws That Run Everything. Joint/marginal/conditional distributions; independence; the Law of Iterated Expectations and Law of Total Variance, with a portfolio-decomposition worked example (Sam). [~5,200 w] - Ch 1.2 — Expectations, Variance, Covariance as Geometry. Linearity of expectation, variance/covariance algebra, correlation as a cosine; covariance matrices previewed. Maya’s two-asset budget. [~4,800 w] - Ch 1.3 — Estimators and Their Sampling Distributions. Estimator vs. estimate; bias, variance, MSE, consistency; the sampling distribution of the mean by simulation. [~5,000 w] - Ch 1.4 — The LLN and the CLT, Shown Not Asserted. Weak LLN; CLT; Monte Carlo convergence plots; when the CLT is slow (heavy tails, crypto returns — Devon). [~5,400 w] - Ch 1.5 — Hypothesis Testing Done Right. Null/alternative, size, power, the t-test derived; confidence intervals as inverted tests; one- vs. two-sided; effect sizes. [~5,600 w]

Daily Problem Sets (≈6 problems each; solutions → Appendix E) - PS 1.1 Conditional probability & LIE/LTV drills [~2,200 w] - PS 1.2 Expectation/variance algebra + correlation geometry [~2,000 w] - PS 1.3 Bias/variance/MSE of estimators (analytic + simulated) [~2,300 w] - PS 1.4 CLT/LLN simulation experiments [~2,400 w] - PS 1.5 Power & size calculations, CI construction [~2,300 w]

Notebooks (one per chapter; markdown + code + “Your Turn”) - nb1.1 LIE/LTV by simulation [~1,400 w / ~20 cells] - nb1.2 Covariance matrices & correlation heatmaps [~1,300 w / ~18 cells] - nb1.3 Sampling-distribution explorer [~1,500 w / ~22 cells] - nb1.4 CLT/LLN convergence animations [~1,600 w / ~24 cells] - nb1.5 Power curves & the t-test from scratch [~1,500 w / ~22 cells]

Lab 1 (Lab Manual): “Build a Coin-Flip Universe.” Simulate sampling distributions, verify CLT, measure empirical size/power of a test you wrote yourself. [~3,400 w]
Mentor Session 1 (Lei Gao): “What is a finding?” — pre-read packet (a 1-page excerpt framing), 3 Socratic warm-ups, 5-slide deck, 3 stretch questions tied to Intraday Momentum (Gao, Han, Li & Zhou 2018), post-session reflection prompt. [~2,200 w]
Assessment W1 + rubric — short-answer + a mini-simulation task; analytic rubric. [~1,800 w]

Week 1 subtotal: ~62,000 words (chapters 26,000 · problems 11,200 · notebooks 7,300 · lab 3,400 · mentor 2,200 · assessment 1,800 · narrative 1,300 · overhead/figures prose ~8,800)

WEEK 2 — The OLS Engine

Theme: Ordinary least squares in matrix form, its guarantees (Gauss–Markov, FWL), and the three ways the textbook story breaks (heteroskedasticity, clustering, misspecification).

W2 Opening narrative — A regression is a machine with knobs; this week we open the case. [~1,300 w]

Chapters (technical week — longer) - Ch 2.1 — OLS in Matrix Form. \mathbf{y}=\mathbf{X}\boldsymbol\beta+\boldsymbol\varepsilon; the normal equations; \hat{\boldsymbol\beta}=(\mathbf{X}'\mathbf{X})^{-1}\mathbf{X}'\mathbf{y} as projection; geometry of the hat matrix. [~6,200 w] - Ch 2.2 — Gauss–Markov and the Meaning of “Best.” The five assumptions stated as testable claims; BLUE; what “efficient” buys you; unbiasedness vs. consistency. [~6,000 w] - Ch 2.3 — The Frisch–Waugh–Lovell Theorem. Partialling-out; controls as residualization; why FWL is the key to reading every multivariate coefficient; demeaning = fixed effects preview. [~6,400 w] - Ch 2.4 — When the Errors Misbehave: Heteroskedasticity, Clustering, HC/HAC. Robust SEs (HC1/HC2/HC3), clustered SEs, Newey–West (HAC); Petersen (2009) panel-SE taxonomy; why your t-stat lies. [~6,800 w] - Ch 2.5 — Misspecification: OVB, Measurement Error, Functional Form. Omitted-variable-bias sign analysis; attenuation from classical measurement error; RESET & functional-form diagnostics; the bias–consistency ledger. [~6,600 w]

Daily Problem Sets - PS 2.1 Matrix-OLS derivations + numerical inversion [~2,400 w] - PS 2.2 Gauss–Markov assumption-breaking diagnostics [~2,300 w] - PS 2.3 FWL by-hand partialling-out exercises [~2,400 w] - PS 2.4 Robust/clustered/HAC SE computation + comparison [~2,500 w] - PS 2.5 OVB sign predictions + measurement-error simulation [~2,400 w]

Notebooks - nb2.1 OLS from scratch with NumPy vs. statsmodels [~1,600 w / ~24 cells] - nb2.2 Gauss–Markov Monte Carlo (efficiency demo) [~1,500 w / ~22 cells] - nb2.3 FWL residualization, visualized [~1,500 w / ~22 cells] - nb2.4 SE flavors on a clustered panel [~1,700 w / ~24 cells] - nb2.5 Biased-estimator lab (simulate OVB & attenuation) [~1,700 w / ~24 cells]

Lab 2 (Lab Manual): “Replicate a Textbook Fama–MacBeth on CRSP.” Two-pass cross-sectional regressions; rolling betas; FM standard errors; pin CRSP snapshot date; licensed-data-on-GMU-infra note. [~4,200 w]
Mentor Session 2 (Lei Gao): “Why your standard error is the whole ballgame.” Pre-read packet, 3 warm-ups, 5-slide deck, 3 stretch questions tied to Petersen (2009) and Gao’s panel work, reflection. [~2,200 w]
Assessment W2 + rubric — derive-and-interpret + a small replication; rubric. [~2,000 w]

Week 2 subtotal: ~68,000 words (chapters 32,000 · problems 12,000 · notebooks 8,000 · lab 4,200 · mentor 2,200 · assessment 2,000 · narrative 1,300 · overhead/figures prose ~6,300)

WEEK 3 — Causal Inference I: Potential Outcomes, Selection-on-Observables, and Instruments

Theme: Correlation is cheap; the Rubin causal model makes “effect” precise, and we earn causal claims through design, not control variables.

W3 Opening narrative — Maya’s loan-approval puzzle: why “controlling for income” isn’t enough. [~1,400 w]

Chapters (technical week — longer) - Ch 3.1 — Potential Outcomes, SUTVA, and the Fundamental Problem. Treatment effects (ATE/ATT/LATE); the missing counterfactual; SUTVA; selection bias decomposed. [~6,400 w] - Ch 3.2 — Selection on Observables I: Matching & Propensity Scores. Conditional independence; nearest-neighbor & caliper matching; PSM; common support; balance diagnostics. [~6,400 w] - Ch 3.3 — Selection on Observables II: Entropy Balancing & Doubly-Robust Estimation. Entropy balancing (Hainmueller); IPW; AIPW/doubly-robust estimators; why “doubly” robust. [~6,600 w] - Ch 3.4 — Instrumental Variables. Relevance & exclusion; 2SLS as FWL-with-an-instrument; LATE interpretation; weak instruments — first-stage F, Stock–Yogo critical values, Olea–Pflueger effective F. [~7,000 w] - Ch 3.5 — Reading IV in the Wild + The Weak-IV Pathology. Anderson–Rubin inference; many-instruments bias; a guided reproduction of a weak-instrument disaster. [~6,400 w]

Daily Problem Sets - PS 3.1 Potential-outcomes algebra; selection-bias decomposition [~2,400 w] - PS 3.2 Hand-matching + propensity-score estimation [~2,500 w] - PS 3.3 Entropy-balancing weights + AIPW by hand [~2,500 w] - PS 3.4 2SLS derivation + weak-IV F diagnostics [~2,600 w] - PS 3.5 Anderson–Rubin CIs; instrument-validity critique [~2,400 w]

Notebooks - nb3.1 Simulating potential outcomes & selection bias [~1,600 w / ~22 cells] - nb3.2 PSM with balance diagnostics [~1,700 w / ~24 cells] - nb3.3 Entropy balancing vs. IPW vs. AIPW [~1,700 w / ~24 cells] - nb3.4 2SLS with linearmodels; first-stage F [~1,700 w / ~24 cells] - nb3.5 Weak-IV pathology reproduction [~1,800 w / ~26 cells]

Lab 3 (Lab Manual): “Reproduce a Weak-IV Pathology.” Build a DGP where 2SLS is badly biased toward OLS; show Olea–Pflueger effective-F flags it; contrast with a strong instrument. [~4,200 w]
Mentor Session 3 (Lei Gao): “Natural experiments: finding the lever nature pulled.” Tied to Deng, Gao & Kim (2020) short-sale natural experiment. Pre-read, 3 warm-ups, 5-slide deck, 3 stretch questions, reflection. [~2,200 w]
Assessment W3 + rubric. [~2,000 w]

Week 3 subtotal: ~70,000 words (chapters 32,800 · problems 12,400 · notebooks 8,500 · lab 4,200 · mentor 2,200 · assessment 2,000 · narrative 1,400 · overhead/figures prose ~6,500)

WEEK 4 — Causal Inference II: DiD, RD, Synthetic Control, and Shift-Share

Theme: The modern panel/quasi-experimental toolkit — and the recent literature showing the “obvious” estimators are often wrong.

W4 Opening narrative — A state changes a law; did it change behavior? Priya’s climate-insurance shock. [~1,400 w]

Chapters (technical week — longer) - Ch 4.1 — Difference-in-Differences & Event Studies. 2×2 DiD; parallel-trends assumption (and how to interrogate it); event-study/leads-and-lags specification; pre-trend tests. [~6,600 w] - Ch 4.2 — The Staggered-Adoption Crisis. Why TWFE is biased under heterogeneous timing: Goodman-Bacon decomposition; the negative-weights problem (de Chaisemartin–D’Haultfœuille); Sun–Abraham, Callaway–Sant’Anna, and Borusyak–Jaravel–Spiess estimators. [~7,200 w] - Ch 4.3 — Regression Discontinuity. Sharp & fuzzy RD; local polynomial estimation; bandwidth choice (Imbens–Kalyanaraman, Calonico–Cattaneo–Titiunik robust bias-corrected); McCrary density test; RD validity threats. [~7,000 w] - Ch 4.4 — Synthetic Control & Synthetic DiD. Abadie’s synthetic control; donor pools & placebo inference; Arkhangelsky et al. synthetic DiD; when one treated unit is all you have. [~6,400 w] - Ch 4.5 — Bartik / Shift-Share Designs. Shift-share instruments; the Borusyak–Hull–Jaravel exposure-weights view; Goldsmith-Pinkham–Sorkin–Swift identification; where the variation really comes from. [~6,400 w]

Daily Problem Sets - PS 4.1 2×2 DiD + event-study construction [~2,400 w] - PS 4.2 Goodman-Bacon decomposition by hand; negative-weights demo [~2,600 w] - PS 4.3 RD local-polynomial fitting + bandwidth experiments [~2,500 w] - PS 4.4 Synthetic control weights + placebo inference [~2,500 w] - PS 4.5 Shift-share exposure weights + identification critique [~2,400 w]

Notebooks - nb4.1 Event-study plots & parallel-trends tests [~1,700 w / ~24 cells] - nb4.2 Staggered DiD: TWFE vs. Callaway–Sant’Anna (pyfixest/differences) [~1,900 w / ~28 cells] - nb4.3 RD with rdrobust (CCT bandwidths) [~1,800 w / ~26 cells] - nb4.4 Synthetic control & synthetic DiD [~1,800 w / ~26 cells] - nb4.5 Bartik instrument decomposition [~1,700 w / ~24 cells]

Lab 4 (Lab Manual): “A Clean DiD on HMDA + a State Policy Shock.” Build a panel from HMDA, define treatment by a state regulatory change, run event-study + Callaway–Sant’Anna, stress parallel trends; ties to the fair-lending thread. [~4,600 w]
Mentor Session 4 (Lei Gao): “Detecting discrimination with a clean design.” Tied to Sun & Gao (2019, PNAS) fair-lending. Pre-read, 3 warm-ups, 5-slide deck, 3 stretch questions, reflection. [~2,200 w]
Assessment W4 + rubric. [~2,000 w]

Week 4 subtotal: ~73,000 words (chapters 33,600 · problems 12,400 · notebooks 8,900 · lab 4,600 · mentor 2,200 · assessment 2,000 · narrative 1,400 · overhead/figures prose ~7,900)

WEEK 5 — Reading the Frontier I: Decoding Empirical Papers

Theme: One paper per day. For each, a 4-page Reader’s Guide with a fixed anatomy: research question · identification strategy · data · table-by-table reading order · what’s clever · what’s vulnerable · three replication exercises.

W5 Opening narrative — How a professional reads a paper: not front-to-back, but tables-first. [~1,400 w]

Chapters (Reader’s Guides, ~4 pp ≈ 2,800–3,400 w each) - Ch 5.1 — Reader’s Guide: Fama & French (1992), “The Cross-Section of Expected Stock Returns,” JF 47(2):427–465. Size/value, sorting, the death of CAPM beta. [~3,200 w] - Ch 5.2 — Reader’s Guide: Fama & French (1993), “Common Risk Factors…,” JFE 33(1):3–56. The factor-model machinery; time-series vs. cross-section. [~3,200 w] - Ch 5.3 — Reader’s Guide: Jegadeesh & Titman (1993), “Returns to Buying Winners and Selling Losers,” JF 48(1):65–91. Momentum portfolios; overlapping holding periods. [~3,100 w] - Ch 5.4 — Reader’s Guide: Petersen (2009), “Estimating Standard Errors in Finance Panel Data Sets,” RFS 22(1):435–480. A methods paper read as a paper; clustering choices. [~3,100 w] - Ch 5.5 — Reader’s Guide: Bertrand, Duflo & Mullainathan (2004), “How Much Should We Trust DiD Estimates?” QJE 119(1):249–275. Serial correlation in DiD; the placebo-law method. [~3,100 w]

Daily Problem Sets (replication-flavored; ~6 tasks each) - PS 5.1 FF92 sorting & univariate portfolio replication tasks [~2,300 w] - PS 5.2 FF93 factor construction & time-series regressions [~2,300 w] - PS 5.3 Momentum portfolio replication + transaction-cost critique [~2,300 w] - PS 5.4 Re-run Petersen’s clustering comparison on a panel [~2,200 w] - PS 5.5 Reproduce a BDM placebo-DiD false-positive [~2,300 w]

Notebooks - nb5.1 FF92 portfolio sorts on CRSP/Compustat [~1,700 w / ~24 cells] - nb5.2 FF93 factor regressions [~1,700 w / ~24 cells] - nb5.3 Momentum strategy backtest [~1,800 w / ~26 cells] - nb5.4 Petersen clustering replication [~1,600 w / ~22 cells] - nb5.5 BDM placebo-law simulation [~1,700 w / ~24 cells]

Reading Guide Pack 5 (meta-guide): the fixed Reader’s-Guide template, a “reading order” heuristic, a glossary of table conventions, and a self-check rubric students apply to each paper. [~3,000 w]
Mentor Session 5 (Lei Gao): “Anatomy of a JF paper.” Tied to the Rainbow of Municipal Credit municipal-borrowing working paper (AEA 2025). Pre-read, 3 warm-ups, 5-slide deck, 3 stretch questions, reflection. [~2,200 w]
Assessment W5 + rubric — students write their own 4-page Reader’s Guide on an unseen paper; rubric. [~2,200 w]

Week 5 subtotal: ~58,000 words (guides 15,700 · problems 11,400 · notebooks 8,500 · reading pack 3,000 · mentor 2,200 · assessment 2,200 · narrative 1,400 · overhead/figures prose ~13,600)

[CHECK] Verify exact page ranges for Fama–French (1992) JF 47(2):427–465; FF (1993) JFE 33(1):3–56; Jegadeesh–Titman (1993) JF 48(1):65–91; Petersen (2009) RFS 22(1):435–480; Bertrand–Duflo–Mullainathan (2004) QJE 119(1):249–275. Years are confident; precise pages to confirm against the journals before publication.

WEEK 6 — Reading the Frontier II: Text, Modern Empirics, and the AI Co-Pilot

Theme: Text-as-data and machine-learning-flavored empirical finance, plus a full module on using LLMs responsibly as a research co-pilot.

W6 Opening narrative — When the data is words: 10-Ks, 8-Ks, and the temptation to let a model do your thinking. [~1,500 w]

Chapters (Reader’s Guides + the AI module) - Ch 6.1 — Reader’s Guide: Kogan, Papanikolaou, Seru & Stoffman (2017), “Technological Innovation, Resource Allocation, and Growth,” QJE 132(2):665–712. Patent-value measure from stock returns. [~3,200 w] - Ch 6.2 — Reader’s Guide: Hoberg & Phillips (2016), “Text-Based Network Industries and Endogenous Product Differentiation,” JPE 124(5):1423–1465. TNIC; cosine similarity of 10-K text. [~3,200 w] - Ch 6.3 — Reader’s Guide: Loughran & McDonald (2011), “When Is a Liability Not a Liability? Textual Analysis…,” JF 66(1):35–65. Finance-specific sentiment dictionaries; the bag-of-words era. [~3,200 w] - Ch 6.4 — Reader’s Guide (paired): Bartlett, Morse, Stanton & Wallace (2022), “Consumer-Lending Discrimination in the FinTech Era,” JFE 143(1):30–56; and Bhutta, Hizmo & Ringo, “How Much Does Racial Bias Affect Mortgage Lending?” Federal Reserve FEDS / working paper. Fair-lending in the algorithmic era; ties to Sun & Gao (2019). [~3,600 w] - Ch 6.5 — The AI Co-Pilot for Research (LLM-in-the-Loop). Prompt patterns for empirical work; RAG over 10-Ks; LLM text classification with out-of-sample validation; critical limits — hallucinated cites, look-ahead/training-data leakage, prompt-induced p-hacking; reproducibility of stochastic outputs. [~7,200 w]

Daily Problem Sets - PS 6.1 KPSS patent-value measure construction tasks [~2,300 w] - PS 6.2 10-K cosine-similarity / TNIC mini-build [~2,400 w] - PS 6.3 LM dictionary sentiment vs. a naive dictionary [~2,300 w] - PS 6.4 Fair-lending decomposition + disparate-impact critique [~2,400 w] - PS 6.5 LLM-classification with held-out validation + leakage audit [~2,600 w]

Notebooks - nb6.1 Building a patent-value panel [~1,700 w / ~24 cells] - nb6.2 10-K text vectorization & similarity [~1,900 w / ~28 cells] - nb6.3 Loughran–McDonald sentiment pipeline [~1,800 w / ~26 cells] - nb6.4 Mortgage-disparity decomposition on HMDA [~1,800 w / ~26 cells] - nb6.5 AI co-pilot lab: Anthropic Messages API, GMU Azure OpenAI deployment (${AZURE_OPENAI_KEY}), local Ollama / Hopper A100 fallback; RAG over 10-Ks; OOS-validated classification [~2,600 w / ~36 cells]

Reading Guide Pack 6 + AI Lab Manual: RAG architecture diagram (prose), prompt-pattern catalog, an evaluation harness for LLM labels (precision/recall vs. hand labels), and a “responsible-use & disclosure” checklist for the capstone. [~4,000 w]
Mentor Session 6 (Lei Gao): “Text as data, and AI without fooling yourself.” Tied to Gao, Han, Kim & Pan (2024, JCF, 84:102520) overlapping supply-chain institutional ownership and supplier-firm earnings management (as a measured construct), and Gao et al. (forthcoming) AI trading simulation. Pre-read, 3 warm-ups, 5-slide deck, 3 stretch questions, reflection. [~2,400 w]
Assessment W6 + rubric — build + validate one LLM text classifier with an OOS report; rubric weights validation & honesty. [~2,400 w]

Week 6 subtotal: ~62,000 words (guides+AI chapter 20,400 · problems 12,000 · notebooks 9,800 · reading/AI pack 4,000 · mentor 2,400 · assessment 2,400 · narrative 1,500 · overhead/figures prose ~9,500)

[CHECK] Confirm exact pages/venues: KPSS (2017) QJE 132(2):665–712; Hoberg–Phillips (2016) JPE 124(5):1423–1465; Loughran–McDonald (2011) JF 66(1):35–65; Bartlett–Morse–Stanton–Wallace (2022) JFE 143(1):30–56. Bhutta–Hizmo–Ringo publication venue/year (Federal Reserve FEDS working paper vs. journal) to be verified before citing a page range.

WEEK 7 — Independent Research Project I: From Question to Pre-Analysis Plan

Theme: Turn a hunch into a falsifiable, pre-registered empirical design with data in hand.

W7 Opening narrative — The scariest blank page in science: choosing your own question. [~1,500 w]

Chapters - Ch 7.1 — Idea-Generation Workshop. From puzzle to testable hypothesis; the “so what / who cares / what’s new” filter; mapping each student’s interest (Maya/Devon/Priya/Sam/Leah) to a feasible dataset. [~5,200 w] - Ch 7.2 — Data Acquisition in Practice. WRDS Cloud, SEC EDGAR (full-text + XBRL), FRED, HMDA, 13F, USPTO PatentsView, yfinance; rate limits, licensing, and building a reproducible pull. [~5,600 w] - Ch 7.3 — The Pre-Analysis Plan (short form, Olken 2015). Hypotheses, primary spec, controls/FEs/clustering, sample, multiple-testing plan (FDR), what would falsify you; registering before you peek. [~5,200 w] - Ch 7.4 — Building the Analysis Dataset. Merging keys (PERMNO/GVKEY/CIK crosswalks), survivorship & look-ahead bias, winsorizing, missing-data discipline, a documented data-build script. [~5,400 w] - Ch 7.5 — Identification Memo. Writing the one-paragraph identifying-assumption statement and the threats-and-responses table that the whole paper will defend. [~4,800 w]

Daily Problem Sets (project-scaffolding deliverables) - PS 7.1 Three candidate questions scored on the feasibility/novelty rubric [~2,000 w] - PS 7.2 A working data-pull script + data card for one source [~2,200 w] - PS 7.3 A complete pre-analysis plan draft [~2,400 w] - PS 7.4 A reproducible merged analysis dataset + diagnostics [~2,300 w] - PS 7.5 Identification memo + threats table [~2,100 w]

Notebooks - nb7.1 Idea-to-spec template notebook [~1,400 w / ~18 cells] - nb7.2 Multi-source data-pull harness [~1,900 w / ~28 cells] - nb7.3 PAP companion (power calc for your design) [~1,600 w / ~22 cells] - nb7.4 Dataset-build & validation notebook [~1,800 w / ~26 cells] - nb7.5 First-look regressions (frozen until PAP filed) [~1,600 w / ~22 cells]

Lab 7 (Lab Manual): “Your Data, Reproducibly.” Stand up a GitHub repo from the template, wire GitHub Classroom, pin environments, write a README + data card, and file the PAP as a tagged commit. [~4,200 w]
Mentor Session 7 (Lei Gao): “How I pick a project — and kill one.” Tied to Gao, He & Wu (2023, JFQA) CSR/price-pressure and the Rainbow of Municipal Credit pipeline. Pre-read, 3 warm-ups, 5-slide deck, 3 stretch questions, reflection. [~2,200 w]
Assessment W7 + rubric — PAP + identification memo graded on the research-design rubric. [~2,200 w]

Week 7 subtotal: ~60,000 words (chapters 26,200 · problems 11,000 · notebooks 8,300 · lab 4,200 · mentor 2,200 · assessment 2,200 · narrative 1,500 · overhead/figures prose ~4,400)

WEEK 8 — Independent Research Project II: Execution, Robustness, Writing, and Defense

Theme: Run it, break it on purpose, write it up to publication standard, and present.

W8 Opening narrative — A result is not a finding until it survives your own attempts to kill it. [~1,500 w]

Chapters - Ch 8.1 — Execution & Specification Curve. Running the pre-registered spec; specification-curve / multiverse analysis; honest deviations log. [~5,400 w] - Ch 8.2 — Robustness & Inference Stress-Tests. Alternative SEs, placebo tests, sensitivity to bandwidth/controls/sample, multiple-testing corrections (Bonferroni/BH-FDR), Oster (2019) δ for selection-on-unobservables. [~5,800 w] - Ch 8.3 — Writing the Empirical Paper. Structure (intro that promises, lit that positions, data, design, results, robustness, conclusion); table & figure craft per Appendix D; the one-sentence contribution. [~5,600 w] - Ch 8.4 — Peer Review & Revision. Refereeing another student’s paper; responding to a referee report; the revise-and-resubmit memo. [~4,800 w] - Ch 8.5 — The 8-Minute Presentation & the Replication Packet. Slide discipline; defending identification under questions; assembling a one-click replication packet (data, code, README, seed). [~4,800 w]

Daily Problem Sets (capstone-production deliverables) - PS 8.1 Specification curve for your main result [~2,200 w] - PS 8.2 A full robustness battery + write-up [~2,400 w] - PS 8.3 Draft intro + results section against the style guide [~2,300 w] - PS 8.4 A referee report on a peer’s paper + your R&R memo [~2,300 w] - PS 8.5 8-minute deck + replication-packet checklist [~2,000 w]

Notebooks - nb8.1 Specification-curve generator [~1,800 w / ~26 cells] - nb8.2 Robustness-battery harness (placebos, Oster δ) [~1,900 w / ~28 cells] - nb8.3 Publication-quality tables/figures (pyfixest/stargazer-style → LaTeX) [~1,800 w / ~26 cells] - nb8.4 Reproducibility-check notebook (fresh-env rerun) [~1,500 w / ~20 cells] - nb8.5 Final manuscript build (Markdown/LaTeX → PDF) [~1,500 w / ~20 cells]

Lab 8 (Lab Manual): “Final Manuscript + Repo + Defense.” Compile the LaTeX paper (AEA template variant) on Overleaf, finalize the GitHub repo with replication packet, dry-run the 8-minute talk. [~4,600 w]
Mentor Session 8 (Lei Gao): “Defending a result: what a referee actually asks.” Tied to Elnahas, Gao, Hossain & Kim (2024, JFQA) disclosure/CEO-politics. Pre-read, 3 warm-ups, 5-slide deck, 3 stretch questions, reflection. [~2,200 w]
Assessment W8 + rubric — the capstone paper + presentation graded on the full research rubric (the program’s terminal assessment). [~2,600 w]

Week 8 subtotal: ~61,000 words (chapters 26,400 · problems 11,200 · notebooks 8,500 · lab 4,600 · mentor 2,200 · assessment 2,600 · narrative 1,500 · overhead/figures prose ~4,000)

APPENDICES

Appendix A — Math Toolkit (just enough)

A.1 Matrix algebra: vectors, matrix multiply, transpose/inverse, rank, positive-definiteness, the spectral/SVD idea at the intuition level. [~4,200 w]
A.2 Optimization: gradients, first/second-order conditions, constrained optimization & Lagrange, the normal equations as an optimization. [~3,800 w]
A.3 Asymptotics: convergence in probability/distribution, Slutsky, continuous-mapping, the delta method, sandwich-variance intuition. [~4,000 w]
A.4 Probability distributions reference (Normal, t, \chi^2, F, Bernoulli/Binomial, Poisson) + when each shows up. [~2,500 w]

Appendix A subtotal: ~14,500 words

Appendix B — Python + LaTeX Setup

B.1 Toolchain: VS Code, conda env (python=3.11 + pinned stack), Git basics. [~3,000 w]
B.2 GitHub & GitHub Classroom workflow for the camp. [~2,200 w]
B.3 WRDS Cloud access + querying licensed data (and the read-only/GMU-infra rules). [~2,800 w]
B.4 GMU Hopper SLURM templates (batch jobs, A100 GPU for the AI module, Ollama). [~2,800 w]
B.5 Overleaf + the AEA LaTeX template variant; BibTeX hygiene. [~2,400 w]

Appendix B subtotal: ~13,200 words

Appendix C — Data Dictionary Master (~30 data cards)

Each card: provider, coverage, key identifiers, access path, license/§-on-GMU note, gotchas, a “first 10 rows” sketch, and which chapter/lab uses it. ~700–950 words each.

CRSP · Compustat (Fundamentals Annual/Quarterly) · IBES · Thomson/SEC 13F · TRACE · Mergent FISD · OptionMetrics · Capital IQ · FRED · Call Reports (FFIEC) · FR Y-9C · SEC EDGAR 10-K/10-Q · SEC EDGAR 8-K · SEC EDGAR DEF 14A · SEC EDGAR 13F · SEC EDGAR N-PORT · SEC EDGAR XBRL Financial Statements · Treasury/FINRA · FDIC · HMDA · CFPB Consumer Complaints · Census · BLS · BEA · USPTO PatentsView · USAspending · NOAA/FEMA · yfinance/Stooq/Tiingo/Alpha Vantage · FOMC text/GDELT · Etherscan/CoinGecko/DeFiLlama · ECB/BIS/IMF/World Bank/OECD.

~33 cards × ~820 w avg ≈ Appendix C subtotal: ~27,000 words

Appendix D — Style Guide for Empirical-Finance Writing

D.1 Table craft: layout, what goes in notes, reporting t-stats/SEs/stars, significant digits. [~3,200 w]
D.2 Reporting regressions: which coefficients, R^2/within-R^2, FE/cluster disclosure, sample lines. [~2,800 w]
D.3 Robustness sections that persuade; the threats-and-responses table. [~2,400 w]
D.4 The replication packet standard (data, code, seed, README, environment file). [~2,400 w]
D.5 Prose style: hedging vs. overclaiming, causal language discipline, citing primary sources. [~2,200 w]

Appendix D subtotal: ~13,000 words

Appendix E — Solutions Manual (every problem fully worked)

8 weeks × 5 problem sets × ~6 problems = ~240 problems, each with a complete worked solution (derivation, code where relevant, interpretation). Counted separately from the problem statements above (those counts were statement + solution as a pair); this appendix is the canonical home of the full worked solutions, organized by week.

Solutions, Weeks 1–2 [~14,000 w]
Solutions, Weeks 3–4 [~15,500 w]
Solutions, Weeks 5–6 [~14,000 w]
Solutions, Weeks 7–8 [~13,500 w]

Appendix E subtotal: ~57,000 words

Appendices total: ~124,700 words

CAPSTONE GALLERY (5 full example papers, 8–12 pp each)

Publication-style worked exemplars students can model their own paper on. Each ~3,000–3,800 words of paper + ~1,000 words of an annotated “how this paper was built” margin commentary.

Capstone 1 — Fair Lending on HMDA. Disparity in mortgage approval/pricing; decomposition + a clean design; ties to Sun & Gao (2019). [~4,600 w]
Capstone 2 — Common Ownership from 13F. Constructing common-ownership measures and a disclosure/competition outcome (a student-track variant of the anchor paper’s earnings-management outcome); ties to Gao, Han, Kim & Pan (2024), JCF, 84:102520. [~4,600 w]
Capstone 3 — Innovation from USPTO PatentsView. Patent-based innovation measure and a firm-outcome event study; ties to KPSS (2017). [~4,400 w]
Capstone 4 — SEC 8-K Text Classification. Classifying 8-K events with OOS-validated text models and a return reaction study; AI-module methods. [~4,600 w]
Capstone 5 — FRED Macro Event Study. A monetary/macro-announcement event study with proper inference. [~4,400 w]

Capstone Gallery subtotal: ~22,600 words

INSTRUCTOR’S MANUAL (separate document)

IM-1 Pacing guide: the 8-week clock, daily blocks, how to compress to ASSIP 2026’s 8-week full-time calendar (Mon-Fri 9-5). [~3,200 w]
IM-2 Grading rubrics (consolidated): problem sets, assessments, capstone paper, presentation. [~3,400 w]
IM-3 Common student pitfalls, by week, with diagnostic questions and fixes. [~3,600 w]
IM-4 Suggested guest lectures & how to slot them; mentor-session facilitation notes. [~2,400 w]
IM-5 Answer keys to assessments + sample graded work (anchor papers for the rubric). [~3,400 w]
IM-6 Equity/access notes: WRDS seats, compute, accommodations; data-licensing compliance. [~2,000 w]

Instructor’s Manual subtotal: ~18,000 words

GRAND TOTAL — Summary Table

Section	Est. words
Front Matter	11,500
Week 1 — Probability, Sampling, Inference	62,000
Week 2 — The OLS Engine	68,000
Week 3 — Causal Inference I (PO, matching, IV)	70,000
Week 4 — Causal Inference II (DiD, RD, SC, shift-share)	73,000
Week 5 — Reading the Frontier I	58,000
Week 6 — Reading the Frontier II + AI Co-Pilot	62,000
Week 7 — Research Project I (question → PAP)	60,000
Week 8 — Research Project II (execution → defense)	61,000
Appendix A — Math Toolkit	14,500
Appendix B — Python + LaTeX Setup	13,200
Appendix C — Data Dictionary (~33 cards)	27,000
Appendix D — Style Guide	13,000
Appendix E — Solutions Manual	57,000
Capstone Gallery (5 papers)	22,600
Instructor’s Manual	18,000
GRAND TOTAL	~690,800

Word-budget note. The line-item estimates above sum to ~690,800 words — above the 300,000–500,000 target. This TOC is intentionally the maximal plan so no deliverable is dropped. The published target is hit by treating the per-chapter prose floor (≈4,000 w) rather than the ceiling, trimming notebook narration to code-comment level, and consolidating solutions, as follows. Authoritative budget to write to:

Section Target words

Front Matter 10,000

Weeks 1–8 (8 × ~33,000 avg) 264,000

Appendices A–D 55,000

Appendix E (Solutions) 40,000

Capstone Gallery 18,000

Instructor’s Manual 13,000

PUBLISHED GRAND TOTAL ~400,000

The book is written to ~400,000 words, squarely inside the 300k–500k range. Each week is budgeted at ~33,000 words: 5 chapters at the ~4,000–5,500 w range (technical Weeks 2–4 at the high end, reading Weeks 5–6 leaner per-guide), 5 problem sets at ~1,800 w each (statement only; full solutions consolidated in Appendix E at ~40,000 w total), 5 notebooks at ~1,200 w narration each, one lab/reading pack (~3,500 w), one mentor session (~2,000 w), one assessment (~1,800 w), and the opening narrative (~1,300 w). Writing slices MUST hit the target column; the maximal column is a ceiling, not a quota.

Section	Target words
Front Matter	10,000
Weeks 1–8 (8 × ~33,000 avg)	264,000
Appendices A–D	55,000
Appendix E (Solutions)	40,000
Capstone Gallery	18,000
Instructor’s Manual	13,000
PUBLISHED GRAND TOTAL	~400,000

Open verification items ([CHECK] consolidated)

Exact page ranges for the Week 5 classics (FF92, FF93, JT93, Petersen 2009, BDM 2004) — years confident, pages to confirm.
Week 6 modern papers’ exact pages/venues (KPSS 2017, Hoberg–Phillips 2016, Loughran–McDonald 2011, Bartlett et al. 2022) and Bhutta–Hizmo–Ringo venue/year (Fed FEDS working paper vs. journal).
All Gao citations are used verbatim from CONVENTIONS.md §6 / spec §5 and are not flagged.
The POM supply-chain-risk → bank-loan-cost manuscript (ms POM-Nov-25-OA-1924) is not used in this TOC (not in the CV selected list per spec §5); excluded until confirmed.

# Table of Contents — *Empirical Finance from the Ground Up* ### A Textbook, Lab Manual, and Problem-Set Bank for the GMU 8-Week Empirical-Finance Research Camp **Host:** Prof. Lei Gao, Associate Professor of Finance, Costello College of Business, George Mason University **Feeds:** ASSIP 2026 (Aspiring Scientists Summer Internship Program) (GMU College of Science) **Audience:** mathematically gifted U.S. high-school students (AP Calc BC + AP Stats; some Python) > **How to read this TOC.** Every line item carries a word-count estimate `[~N w]`. Prose chapters > count finished prose only (worked examples included; code listings counted as their explanatory > prose). Problem sets count problem statements **plus** the fully worked solutions that live in > Appendix E. Notebooks count their markdown narration + code comments + "Your Turn" prose (cell > counts shown separately, e.g. `~22 cells`). Per-week subtotals and a grand total appear at the end. > Citations follow `CONVENTIONS.md §6`; anything unverified is tagged **[CHECK]**. --- ## FRONT MATTER | # | Item | Description | Est. words | |---|------|-------------|-----------| | FM-0 | **Title page, copyright, dedication** | GMU / Costello / College of Science branding; CC-BY-NC license note | 300 | | FM-1 | **Preface** (by Prof. Gao) | Why empirical finance, why high-schoolers can do real research, what the camp feeds into; grounded in verified CV facts | 1,700 | | FM-2 | **Articulation Matrix** (Camp ↔ ASSIP 2026) | 2-page matrix: 8-week camp vs. the real ASSIP 2026 program (8 wks full-time (Mon-Fri 9-5), Mon-Fri 9-5, Jun 22 - Aug 12, 2026; CRSP/Compustat/EDGAR/Bloomberg; Claude/ChatGPT; Journal of Student-Scientists' Research (JSSR), GMU Mason Publishing + JSSR (Journal of Student-Scientists' Research); 3 GMU College of Science credits + \$1,299 tuition (full waivers for Pell-eligible / reduced-lunch students)). Shows where the camp **deepens** vs. where ASSIP 2026 **leads** | 1,600 | | FM-3 | **How to Use This Book** | Daily rhythm; how chapters/problems/notebooks/labs/mentor sessions interlock; the reveal-the-trick structure; the recurring student cast (Maya, Devon, Priya, Sam, Leah) | 1,800 | | FM-4 | **Prerequisite Self-Test** | 20 questions (calculus, probability/stats, light Python) + full solutions + routing table ("missed these → read Appendix A/B first") | 4,500 | | FM-5 | **Notation & Conventions quick-card** | One-page reproduction of the `CONVENTIONS.md` notation table + spec-discipline checklist | 700 | | FM-6 | **Roadmap diagram + dependency graph (prose)** | Narrative walkthrough of the 8-week arc and how skills compound | 900 | **Front Matter subtotal: ~11,500 words** --- # THE 8-WEEK ARC > Each week contains: (1) a week-opening narrative, (2) **five** chapters, (3) **five** daily problem > sets (~6 problems each; solutions in Appendix E), (4) a **notebook per chapter** with a "Your Turn" > extension, (5) a lab manual (Weeks 1–4, 7–8) or reading guides (Weeks 5–6), (6) a 60-minute **Lei > Gao mentor session**, and (7) an **end-of-week assessment with rubric**. --- ## WEEK 1 — Probability, Sampling, and the Logic of Inference **Theme:** Rebuild probability and frequentist inference from the ground up, using simulation as the microscope, so OLS in Week 2 rests on solid foundations. - **W1 Opening narrative** — From "the average return was 8%" to "what could we have seen instead?": randomness as the object of study. `[~1,300 w]` **Chapters** - **Ch 1.1 — Joint, Conditional, and the Two Laws That Run Everything.** Joint/marginal/conditional distributions; independence; the Law of Iterated Expectations and Law of Total Variance, with a portfolio-decomposition worked example (Sam). `[~5,200 w]` - **Ch 1.2 — Expectations, Variance, Covariance as Geometry.** Linearity of expectation, variance/covariance algebra, correlation as a cosine; covariance matrices previewed. Maya's two-asset budget. `[~4,800 w]` - **Ch 1.3 — Estimators and Their Sampling Distributions.** Estimator vs. estimate; bias, variance, MSE, consistency; the sampling distribution of the mean by simulation. `[~5,000 w]` - **Ch 1.4 — The LLN and the CLT, Shown Not Asserted.** Weak LLN; CLT; Monte Carlo convergence plots; when the CLT is slow (heavy tails, crypto returns — Devon). `[~5,400 w]` - **Ch 1.5 — Hypothesis Testing Done Right.** Null/alternative, size, power, the t-test derived; confidence intervals as inverted tests; one- vs. two-sided; effect sizes. `[~5,600 w]` **Daily Problem Sets** (≈6 problems each; solutions → Appendix E) - **PS 1.1** Conditional probability & LIE/LTV drills `[~2,200 w]` - **PS 1.2** Expectation/variance algebra + correlation geometry `[~2,000 w]` - **PS 1.3** Bias/variance/MSE of estimators (analytic + simulated) `[~2,300 w]` - **PS 1.4** CLT/LLN simulation experiments `[~2,400 w]` - **PS 1.5** Power & size calculations, CI construction `[~2,300 w]` **Notebooks** (one per chapter; markdown + code + "Your Turn") - **nb1.1** LIE/LTV by simulation `[~1,400 w / ~20 cells]` - **nb1.2** Covariance matrices & correlation heatmaps `[~1,300 w / ~18 cells]` - **nb1.3** Sampling-distribution explorer `[~1,500 w / ~22 cells]` - **nb1.4** CLT/LLN convergence animations `[~1,600 w / ~24 cells]` - **nb1.5** Power curves & the t-test from scratch `[~1,500 w / ~22 cells]` - **Lab 1 (Lab Manual): "Build a Coin-Flip Universe."** Simulate sampling distributions, verify CLT, measure empirical size/power of a test you wrote yourself. `[~3,400 w]` - **Mentor Session 1 (Lei Gao):** *"What is a finding?"* — pre-read packet (a 1-page excerpt framing), 3 Socratic warm-ups, 5-slide deck, 3 stretch questions tied to Intraday Momentum (Gao, Han, Li & Zhou 2018), post-session reflection prompt. `[~2,200 w]` - **Assessment W1 + rubric** — short-answer + a mini-simulation task; analytic rubric. `[~1,800 w]` **Week 1 subtotal: ~62,000 words** *(chapters 26,000 · problems 11,200 · notebooks 7,300 · lab 3,400 · mentor 2,200 · assessment 1,800 · narrative 1,300 · overhead/figures prose ~8,800)* --- ## WEEK 2 — The OLS Engine **Theme:** Ordinary least squares in matrix form, its guarantees (Gauss–Markov, FWL), and the three ways the textbook story breaks (heteroskedasticity, clustering, misspecification). - **W2 Opening narrative** — A regression is a machine with knobs; this week we open the case. `[~1,300 w]` **Chapters** (technical week — longer) - **Ch 2.1 — OLS in Matrix Form.** $\mathbf{y}=\mathbf{X}\boldsymbol\beta+\boldsymbol\varepsilon$; the normal equations; $\hat{\boldsymbol\beta}=(\mathbf{X}'\mathbf{X})^{-1}\mathbf{X}'\mathbf{y}$ as projection; geometry of the hat matrix. `[~6,200 w]` - **Ch 2.2 — Gauss–Markov and the Meaning of "Best."** The five assumptions stated as testable claims; BLUE; what "efficient" buys you; unbiasedness vs. consistency. `[~6,000 w]` - **Ch 2.3 — The Frisch–Waugh–Lovell Theorem.** Partialling-out; controls as residualization; why FWL is the key to reading every multivariate coefficient; demeaning = fixed effects preview. `[~6,400 w]` - **Ch 2.4 — When the Errors Misbehave: Heteroskedasticity, Clustering, HC/HAC.** Robust SEs (HC1/HC2/HC3), clustered SEs, Newey–West (HAC); Petersen (2009) panel-SE taxonomy; why your t-stat lies. `[~6,800 w]` - **Ch 2.5 — Misspecification: OVB, Measurement Error, Functional Form.** Omitted-variable-bias sign analysis; attenuation from classical measurement error; RESET & functional-form diagnostics; the bias–consistency ledger. `[~6,600 w]` **Daily Problem Sets** - **PS 2.1** Matrix-OLS derivations + numerical inversion `[~2,400 w]` - **PS 2.2** Gauss–Markov assumption-breaking diagnostics `[~2,300 w]` - **PS 2.3** FWL by-hand partialling-out exercises `[~2,400 w]` - **PS 2.4** Robust/clustered/HAC SE computation + comparison `[~2,500 w]` - **PS 2.5** OVB sign predictions + measurement-error simulation `[~2,400 w]` **Notebooks** - **nb2.1** OLS from scratch with NumPy vs. statsmodels `[~1,600 w / ~24 cells]` - **nb2.2** Gauss–Markov Monte Carlo (efficiency demo) `[~1,500 w / ~22 cells]` - **nb2.3** FWL residualization, visualized `[~1,500 w / ~22 cells]` - **nb2.4** SE flavors on a clustered panel `[~1,700 w / ~24 cells]` - **nb2.5** Biased-estimator lab (simulate OVB & attenuation) `[~1,700 w / ~24 cells]` - **Lab 2 (Lab Manual): "Replicate a Textbook Fama–MacBeth on CRSP."** Two-pass cross-sectional regressions; rolling betas; FM standard errors; pin CRSP snapshot date; licensed-data-on-GMU-infra note. `[~4,200 w]` - **Mentor Session 2 (Lei Gao):** *"Why your standard error is the whole ballgame."* Pre-read packet, 3 warm-ups, 5-slide deck, 3 stretch questions tied to Petersen (2009) and Gao's panel work, reflection. `[~2,200 w]` - **Assessment W2 + rubric** — derive-and-interpret + a small replication; rubric. `[~2,000 w]` **Week 2 subtotal: ~68,000 words** *(chapters 32,000 · problems 12,000 · notebooks 8,000 · lab 4,200 · mentor 2,200 · assessment 2,000 · narrative 1,300 · overhead/figures prose ~6,300)* --- ## WEEK 3 — Causal Inference I: Potential Outcomes, Selection-on-Observables, and Instruments **Theme:** Correlation is cheap; the Rubin causal model makes "effect" precise, and we earn causal claims through design, not control variables. - **W3 Opening narrative** — Maya's loan-approval puzzle: why "controlling for income" isn't enough. `[~1,400 w]` **Chapters** (technical week — longer) - **Ch 3.1 — Potential Outcomes, SUTVA, and the Fundamental Problem.** Treatment effects (ATE/ATT/LATE); the missing counterfactual; SUTVA; selection bias decomposed. `[~6,400 w]` - **Ch 3.2 — Selection on Observables I: Matching & Propensity Scores.** Conditional independence; nearest-neighbor & caliper matching; PSM; common support; balance diagnostics. `[~6,400 w]` - **Ch 3.3 — Selection on Observables II: Entropy Balancing & Doubly-Robust Estimation.** Entropy balancing (Hainmueller); IPW; AIPW/doubly-robust estimators; why "doubly" robust. `[~6,600 w]` - **Ch 3.4 — Instrumental Variables.** Relevance & exclusion; 2SLS as FWL-with-an-instrument; LATE interpretation; weak instruments — first-stage F, Stock–Yogo critical values, Olea–Pflueger effective F. `[~7,000 w]` - **Ch 3.5 — Reading IV in the Wild + The Weak-IV Pathology.** Anderson–Rubin inference; many-instruments bias; a guided reproduction of a weak-instrument disaster. `[~6,400 w]` **Daily Problem Sets** - **PS 3.1** Potential-outcomes algebra; selection-bias decomposition `[~2,400 w]` - **PS 3.2** Hand-matching + propensity-score estimation `[~2,500 w]` - **PS 3.3** Entropy-balancing weights + AIPW by hand `[~2,500 w]` - **PS 3.4** 2SLS derivation + weak-IV F diagnostics `[~2,600 w]` - **PS 3.5** Anderson–Rubin CIs; instrument-validity critique `[~2,400 w]` **Notebooks** - **nb3.1** Simulating potential outcomes & selection bias `[~1,600 w / ~22 cells]` - **nb3.2** PSM with balance diagnostics `[~1,700 w / ~24 cells]` - **nb3.3** Entropy balancing vs. IPW vs. AIPW `[~1,700 w / ~24 cells]` - **nb3.4** 2SLS with `linearmodels`; first-stage F `[~1,700 w / ~24 cells]` - **nb3.5** Weak-IV pathology reproduction `[~1,800 w / ~26 cells]` - **Lab 3 (Lab Manual): "Reproduce a Weak-IV Pathology."** Build a DGP where 2SLS is badly biased toward OLS; show Olea–Pflueger effective-F flags it; contrast with a strong instrument. `[~4,200 w]` - **Mentor Session 3 (Lei Gao):** *"Natural experiments: finding the lever nature pulled."* Tied to Deng, Gao & Kim (2020) short-sale natural experiment. Pre-read, 3 warm-ups, 5-slide deck, 3 stretch questions, reflection. `[~2,200 w]` - **Assessment W3 + rubric.** `[~2,000 w]` **Week 3 subtotal: ~70,000 words** *(chapters 32,800 · problems 12,400 · notebooks 8,500 · lab 4,200 · mentor 2,200 · assessment 2,000 · narrative 1,400 · overhead/figures prose ~6,500)* --- ## WEEK 4 — Causal Inference II: DiD, RD, Synthetic Control, and Shift-Share **Theme:** The modern panel/quasi-experimental toolkit — and the recent literature showing the "obvious" estimators are often wrong. - **W4 Opening narrative** — A state changes a law; did it change behavior? Priya's climate-insurance shock. `[~1,400 w]` **Chapters** (technical week — longer) - **Ch 4.1 — Difference-in-Differences & Event Studies.** 2×2 DiD; parallel-trends assumption (and how to interrogate it); event-study/leads-and-lags specification; pre-trend tests. `[~6,600 w]` - **Ch 4.2 — The Staggered-Adoption Crisis.** Why TWFE is biased under heterogeneous timing: Goodman-Bacon decomposition; the negative-weights problem (de Chaisemartin–D'Haultfœuille); Sun–Abraham, Callaway–Sant'Anna, and Borusyak–Jaravel–Spiess estimators. `[~7,200 w]` - **Ch 4.3 — Regression Discontinuity.** Sharp & fuzzy RD; local polynomial estimation; bandwidth choice (Imbens–Kalyanaraman, Calonico–Cattaneo–Titiunik robust bias-corrected); McCrary density test; RD validity threats. `[~7,000 w]` - **Ch 4.4 — Synthetic Control & Synthetic DiD.** Abadie's synthetic control; donor pools & placebo inference; Arkhangelsky et al. synthetic DiD; when one treated unit is all you have. `[~6,400 w]` - **Ch 4.5 — Bartik / Shift-Share Designs.** Shift-share instruments; the Borusyak–Hull–Jaravel exposure-weights view; Goldsmith-Pinkham–Sorkin–Swift identification; where the variation really comes from. `[~6,400 w]` **Daily Problem Sets** - **PS 4.1** 2×2 DiD + event-study construction `[~2,400 w]` - **PS 4.2** Goodman-Bacon decomposition by hand; negative-weights demo `[~2,600 w]` - **PS 4.3** RD local-polynomial fitting + bandwidth experiments `[~2,500 w]` - **PS 4.4** Synthetic control weights + placebo inference `[~2,500 w]` - **PS 4.5** Shift-share exposure weights + identification critique `[~2,400 w]` **Notebooks** - **nb4.1** Event-study plots & parallel-trends tests `[~1,700 w / ~24 cells]` - **nb4.2** Staggered DiD: TWFE vs. Callaway–Sant'Anna (`pyfixest`/`differences`) `[~1,900 w / ~28 cells]` - **nb4.3** RD with `rdrobust` (CCT bandwidths) `[~1,800 w / ~26 cells]` - **nb4.4** Synthetic control & synthetic DiD `[~1,800 w / ~26 cells]` - **nb4.5** Bartik instrument decomposition `[~1,700 w / ~24 cells]` - **Lab 4 (Lab Manual): "A Clean DiD on HMDA + a State Policy Shock."** Build a panel from HMDA, define treatment by a state regulatory change, run event-study + Callaway–Sant'Anna, stress parallel trends; ties to the fair-lending thread. `[~4,600 w]` - **Mentor Session 4 (Lei Gao):** *"Detecting discrimination with a clean design."* Tied to Sun & Gao (2019, *PNAS*) fair-lending. Pre-read, 3 warm-ups, 5-slide deck, 3 stretch questions, reflection. `[~2,200 w]` - **Assessment W4 + rubric.** `[~2,000 w]` **Week 4 subtotal: ~73,000 words** *(chapters 33,600 · problems 12,400 · notebooks 8,900 · lab 4,600 · mentor 2,200 · assessment 2,000 · narrative 1,400 · overhead/figures prose ~7,900)* --- ## WEEK 5 — Reading the Frontier I: Decoding Empirical Papers **Theme:** One paper per day. For each, a 4-page **Reader's Guide** with a fixed anatomy: research question · identification strategy · data · table-by-table reading order · what's clever · what's vulnerable · three replication exercises. - **W5 Opening narrative** — How a professional reads a paper: not front-to-back, but tables-first. `[~1,400 w]` **Chapters (Reader's Guides, ~4 pp ≈ 2,800–3,400 w each)** - **Ch 5.1 — Reader's Guide: Fama & French (1992), "The Cross-Section of Expected Stock Returns," *JF* 47(2):427–465.** Size/value, sorting, the death of CAPM beta. `[~3,200 w]` - **Ch 5.2 — Reader's Guide: Fama & French (1993), "Common Risk Factors...," *JFE* 33(1):3–56.** The factor-model machinery; time-series vs. cross-section. `[~3,200 w]` - **Ch 5.3 — Reader's Guide: Jegadeesh & Titman (1993), "Returns to Buying Winners and Selling Losers," *JF* 48(1):65–91.** Momentum portfolios; overlapping holding periods. `[~3,100 w]` - **Ch 5.4 — Reader's Guide: Petersen (2009), "Estimating Standard Errors in Finance Panel Data Sets," *RFS* 22(1):435–480.** A methods paper read as a paper; clustering choices. `[~3,100 w]` - **Ch 5.5 — Reader's Guide: Bertrand, Duflo & Mullainathan (2004), "How Much Should We Trust DiD Estimates?" *QJE* 119(1):249–275.** Serial correlation in DiD; the placebo-law method. `[~3,100 w]` **Daily Problem Sets** (replication-flavored; ~6 tasks each) - **PS 5.1** FF92 sorting & univariate portfolio replication tasks `[~2,300 w]` - **PS 5.2** FF93 factor construction & time-series regressions `[~2,300 w]` - **PS 5.3** Momentum portfolio replication + transaction-cost critique `[~2,300 w]` - **PS 5.4** Re-run Petersen's clustering comparison on a panel `[~2,200 w]` - **PS 5.5** Reproduce a BDM placebo-DiD false-positive `[~2,300 w]` **Notebooks** - **nb5.1** FF92 portfolio sorts on CRSP/Compustat `[~1,700 w / ~24 cells]` - **nb5.2** FF93 factor regressions `[~1,700 w / ~24 cells]` - **nb5.3** Momentum strategy backtest `[~1,800 w / ~26 cells]` - **nb5.4** Petersen clustering replication `[~1,600 w / ~22 cells]` - **nb5.5** BDM placebo-law simulation `[~1,700 w / ~24 cells]` - **Reading Guide Pack 5 (meta-guide):** the fixed Reader's-Guide template, a "reading order" heuristic, a glossary of table conventions, and a self-check rubric students apply to each paper. `[~3,000 w]` - **Mentor Session 5 (Lei Gao):** *"Anatomy of a JF paper."* Tied to the *Rainbow of Municipal Credit* municipal-borrowing working paper (AEA 2025). Pre-read, 3 warm-ups, 5-slide deck, 3 stretch questions, reflection. `[~2,200 w]` - **Assessment W5 + rubric** — students write their own 4-page Reader's Guide on an unseen paper; rubric. `[~2,200 w]` **Week 5 subtotal: ~58,000 words** *(guides 15,700 · problems 11,400 · notebooks 8,500 · reading pack 3,000 · mentor 2,200 · assessment 2,200 · narrative 1,400 · overhead/figures prose ~13,600)* > **[CHECK]** Verify exact page ranges for Fama–French (1992) *JF* 47(2):427–465; FF (1993) *JFE* 33(1):3–56; Jegadeesh–Titman (1993) *JF* 48(1):65–91; Petersen (2009) *RFS* 22(1):435–480; Bertrand–Duflo–Mullainathan (2004) *QJE* 119(1):249–275. Years are confident; precise pages to confirm against the journals before publication. --- ## WEEK 6 — Reading the Frontier II: Text, Modern Empirics, and the AI Co-Pilot **Theme:** Text-as-data and machine-learning-flavored empirical finance, plus a full module on using LLMs *responsibly* as a research co-pilot. - **W6 Opening narrative** — When the data is words: 10-Ks, 8-Ks, and the temptation to let a model do your thinking. `[~1,500 w]` **Chapters (Reader's Guides + the AI module)** - **Ch 6.1 — Reader's Guide: Kogan, Papanikolaou, Seru & Stoffman (2017), "Technological Innovation, Resource Allocation, and Growth," *QJE* 132(2):665–712.** Patent-value measure from stock returns. `[~3,200 w]` - **Ch 6.2 — Reader's Guide: Hoberg & Phillips (2016), "Text-Based Network Industries and Endogenous Product Differentiation," *JPE* 124(5):1423–1465.** TNIC; cosine similarity of 10-K text. `[~3,200 w]` - **Ch 6.3 — Reader's Guide: Loughran & McDonald (2011), "When Is a Liability Not a Liability? Textual Analysis...," *JF* 66(1):35–65.** Finance-specific sentiment dictionaries; the bag-of-words era. `[~3,200 w]` - **Ch 6.4 — Reader's Guide (paired): Bartlett, Morse, Stanton & Wallace (2022), "Consumer-Lending Discrimination in the FinTech Era," *JFE* 143(1):30–56; and Bhutta, Hizmo & Ringo, "How Much Does Racial Bias Affect Mortgage Lending?" Federal Reserve FEDS / working paper.** Fair-lending in the algorithmic era; ties to Sun & Gao (2019). `[~3,600 w]` - **Ch 6.5 — The AI Co-Pilot for Research (LLM-in-the-Loop).** Prompt patterns for empirical work; RAG over 10-Ks; LLM text classification with **out-of-sample validation**; critical limits — hallucinated cites, look-ahead/training-data leakage, prompt-induced p-hacking; reproducibility of stochastic outputs. `[~7,200 w]` **Daily Problem Sets** - **PS 6.1** KPSS patent-value measure construction tasks `[~2,300 w]` - **PS 6.2** 10-K cosine-similarity / TNIC mini-build `[~2,400 w]` - **PS 6.3** LM dictionary sentiment vs. a naive dictionary `[~2,300 w]` - **PS 6.4** Fair-lending decomposition + disparate-impact critique `[~2,400 w]` - **PS 6.5** LLM-classification with held-out validation + leakage audit `[~2,600 w]` **Notebooks** - **nb6.1** Building a patent-value panel `[~1,700 w / ~24 cells]` - **nb6.2** 10-K text vectorization & similarity `[~1,900 w / ~28 cells]` - **nb6.3** Loughran–McDonald sentiment pipeline `[~1,800 w / ~26 cells]` - **nb6.4** Mortgage-disparity decomposition on HMDA `[~1,800 w / ~26 cells]` - **nb6.5** AI co-pilot lab: **Anthropic Messages API**, **GMU Azure OpenAI** deployment (`${AZURE_OPENAI_KEY}`), local **Ollama / Hopper A100** fallback; RAG over 10-Ks; OOS-validated classification `[~2,600 w / ~36 cells]` - **Reading Guide Pack 6 + AI Lab Manual:** RAG architecture diagram (prose), prompt-pattern catalog, an evaluation harness for LLM labels (precision/recall vs. hand labels), and a "responsible-use & disclosure" checklist for the capstone. `[~4,000 w]` - **Mentor Session 6 (Lei Gao):** *"Text as data, and AI without fooling yourself."* Tied to Gao, Han, Kim & Pan (2024, *JCF*, 84:102520) overlapping supply-chain institutional ownership and supplier-firm earnings management (as a measured construct), and Gao et al. (forthcoming) AI trading simulation. Pre-read, 3 warm-ups, 5-slide deck, 3 stretch questions, reflection. `[~2,400 w]` - **Assessment W6 + rubric** — build + validate one LLM text classifier with an OOS report; rubric weights validation & honesty. `[~2,400 w]` **Week 6 subtotal: ~62,000 words** *(guides+AI chapter 20,400 · problems 12,000 · notebooks 9,800 · reading/AI pack 4,000 · mentor 2,400 · assessment 2,400 · narrative 1,500 · overhead/figures prose ~9,500)* > **[CHECK]** Confirm exact pages/venues: KPSS (2017) *QJE* 132(2):665–712; Hoberg–Phillips (2016) *JPE* 124(5):1423–1465; Loughran–McDonald (2011) *JF* 66(1):35–65; Bartlett–Morse–Stanton–Wallace (2022) *JFE* 143(1):30–56. **Bhutta–Hizmo–Ringo** publication venue/year (Federal Reserve FEDS working paper vs. journal) to be verified before citing a page range. --- ## WEEK 7 — Independent Research Project I: From Question to Pre-Analysis Plan **Theme:** Turn a hunch into a falsifiable, pre-registered empirical design with data in hand. - **W7 Opening narrative** — The scariest blank page in science: choosing your own question. `[~1,500 w]` **Chapters** - **Ch 7.1 — Idea-Generation Workshop.** From puzzle to testable hypothesis; the "so what / who cares / what's new" filter; mapping each student's interest (Maya/Devon/Priya/Sam/Leah) to a feasible dataset. `[~5,200 w]` - **Ch 7.2 — Data Acquisition in Practice.** WRDS Cloud, SEC EDGAR (full-text + XBRL), FRED, HMDA, 13F, USPTO PatentsView, yfinance; rate limits, licensing, and building a reproducible pull. `[~5,600 w]` - **Ch 7.3 — The Pre-Analysis Plan (short form, Olken 2015).** Hypotheses, primary spec, controls/FEs/clustering, sample, multiple-testing plan (FDR), what would falsify you; registering before you peek. `[~5,200 w]` - **Ch 7.4 — Building the Analysis Dataset.** Merging keys (PERMNO/GVKEY/CIK crosswalks), survivorship & look-ahead bias, winsorizing, missing-data discipline, a documented data-build script. `[~5,400 w]` - **Ch 7.5 — Identification Memo.** Writing the one-paragraph identifying-assumption statement and the threats-and-responses table that the whole paper will defend. `[~4,800 w]` **Daily Problem Sets** (project-scaffolding deliverables) - **PS 7.1** Three candidate questions scored on the feasibility/novelty rubric `[~2,000 w]` - **PS 7.2** A working data-pull script + data card for one source `[~2,200 w]` - **PS 7.3** A complete pre-analysis plan draft `[~2,400 w]` - **PS 7.4** A reproducible merged analysis dataset + diagnostics `[~2,300 w]` - **PS 7.5** Identification memo + threats table `[~2,100 w]` **Notebooks** - **nb7.1** Idea-to-spec template notebook `[~1,400 w / ~18 cells]` - **nb7.2** Multi-source data-pull harness `[~1,900 w / ~28 cells]` - **nb7.3** PAP companion (power calc for your design) `[~1,600 w / ~22 cells]` - **nb7.4** Dataset-build & validation notebook `[~1,800 w / ~26 cells]` - **nb7.5** First-look regressions (frozen until PAP filed) `[~1,600 w / ~22 cells]` - **Lab 7 (Lab Manual): "Your Data, Reproducibly."** Stand up a GitHub repo from the template, wire GitHub Classroom, pin environments, write a README + data card, and file the PAP as a tagged commit. `[~4,200 w]` - **Mentor Session 7 (Lei Gao):** *"How I pick a project — and kill one."* Tied to Gao, He & Wu (2023, *JFQA*) CSR/price-pressure and the *Rainbow of Municipal Credit* pipeline. Pre-read, 3 warm-ups, 5-slide deck, 3 stretch questions, reflection. `[~2,200 w]` - **Assessment W7 + rubric** — PAP + identification memo graded on the research-design rubric. `[~2,200 w]` **Week 7 subtotal: ~60,000 words** *(chapters 26,200 · problems 11,000 · notebooks 8,300 · lab 4,200 · mentor 2,200 · assessment 2,200 · narrative 1,500 · overhead/figures prose ~4,400)* --- ## WEEK 8 — Independent Research Project II: Execution, Robustness, Writing, and Defense **Theme:** Run it, break it on purpose, write it up to publication standard, and present. - **W8 Opening narrative** — A result is not a finding until it survives your own attempts to kill it. `[~1,500 w]` **Chapters** - **Ch 8.1 — Execution & Specification Curve.** Running the pre-registered spec; specification-curve / multiverse analysis; honest deviations log. `[~5,400 w]` - **Ch 8.2 — Robustness & Inference Stress-Tests.** Alternative SEs, placebo tests, sensitivity to bandwidth/controls/sample, multiple-testing corrections (Bonferroni/BH-FDR), Oster (2019) δ for selection-on-unobservables. `[~5,800 w]` - **Ch 8.3 — Writing the Empirical Paper.** Structure (intro that promises, lit that positions, data, design, results, robustness, conclusion); table & figure craft per Appendix D; the one-sentence contribution. `[~5,600 w]` - **Ch 8.4 — Peer Review & Revision.** Refereeing another student's paper; responding to a referee report; the revise-and-resubmit memo. `[~4,800 w]` - **Ch 8.5 — The 8-Minute Presentation & the Replication Packet.** Slide discipline; defending identification under questions; assembling a one-click replication packet (data, code, README, seed). `[~4,800 w]` **Daily Problem Sets** (capstone-production deliverables) - **PS 8.1** Specification curve for your main result `[~2,200 w]` - **PS 8.2** A full robustness battery + write-up `[~2,400 w]` - **PS 8.3** Draft intro + results section against the style guide `[~2,300 w]` - **PS 8.4** A referee report on a peer's paper + your R&R memo `[~2,300 w]` - **PS 8.5** 8-minute deck + replication-packet checklist `[~2,000 w]` **Notebooks** - **nb8.1** Specification-curve generator `[~1,800 w / ~26 cells]` - **nb8.2** Robustness-battery harness (placebos, Oster δ) `[~1,900 w / ~28 cells]` - **nb8.3** Publication-quality tables/figures (`pyfixest`/`stargazer`-style → LaTeX) `[~1,800 w / ~26 cells]` - **nb8.4** Reproducibility-check notebook (fresh-env rerun) `[~1,500 w / ~20 cells]` - **nb8.5** Final manuscript build (Markdown/LaTeX → PDF) `[~1,500 w / ~20 cells]` - **Lab 8 (Lab Manual): "Final Manuscript + Repo + Defense."** Compile the LaTeX paper (AEA template variant) on Overleaf, finalize the GitHub repo with replication packet, dry-run the 8-minute talk. `[~4,600 w]` - **Mentor Session 8 (Lei Gao):** *"Defending a result: what a referee actually asks."* Tied to Elnahas, Gao, Hossain & Kim (2024, *JFQA*) disclosure/CEO-politics. Pre-read, 3 warm-ups, 5-slide deck, 3 stretch questions, reflection. `[~2,200 w]` - **Assessment W8 + rubric** — the capstone paper + presentation graded on the full research rubric (the program's terminal assessment). `[~2,600 w]` **Week 8 subtotal: ~61,000 words** *(chapters 26,400 · problems 11,200 · notebooks 8,500 · lab 4,600 · mentor 2,200 · assessment 2,600 · narrative 1,500 · overhead/figures prose ~4,000)* --- # APPENDICES ## Appendix A — Math Toolkit (just enough) - **A.1** Matrix algebra: vectors, matrix multiply, transpose/inverse, rank, positive-definiteness, the spectral/SVD idea at the intuition level. `[~4,200 w]` - **A.2** Optimization: gradients, first/second-order conditions, constrained optimization & Lagrange, the normal equations as an optimization. `[~3,800 w]` - **A.3** Asymptotics: convergence in probability/distribution, Slutsky, continuous-mapping, the delta method, sandwich-variance intuition. `[~4,000 w]` - **A.4** Probability distributions reference (Normal, t, $\chi^2$, F, Bernoulli/Binomial, Poisson) + when each shows up. `[~2,500 w]` **Appendix A subtotal: ~14,500 words** ## Appendix B — Python + LaTeX Setup - **B.1** Toolchain: VS Code, conda env (`python=3.11` + pinned stack), Git basics. `[~3,000 w]` - **B.2** GitHub & GitHub Classroom workflow for the camp. `[~2,200 w]` - **B.3** WRDS Cloud access + querying licensed data (and the read-only/GMU-infra rules). `[~2,800 w]` - **B.4** GMU Hopper SLURM templates (batch jobs, A100 GPU for the AI module, Ollama). `[~2,800 w]` - **B.5** Overleaf + the AEA LaTeX template variant; BibTeX hygiene. `[~2,400 w]` **Appendix B subtotal: ~13,200 words** ## Appendix C — Data Dictionary Master (~30 data cards) > Each card: provider, coverage, key identifiers, access path, license/§-on-GMU note, gotchas, a "first 10 rows" sketch, and which chapter/lab uses it. ~700–950 words each. CRSP · Compustat (Fundamentals Annual/Quarterly) · IBES · Thomson/SEC 13F · TRACE · Mergent FISD · OptionMetrics · Capital IQ · FRED · Call Reports (FFIEC) · FR Y-9C · SEC EDGAR 10-K/10-Q · SEC EDGAR 8-K · SEC EDGAR DEF 14A · SEC EDGAR 13F · SEC EDGAR N-PORT · SEC EDGAR XBRL Financial Statements · Treasury/FINRA · FDIC · HMDA · CFPB Consumer Complaints · Census · BLS · BEA · USPTO PatentsView · USAspending · NOAA/FEMA · yfinance/Stooq/Tiingo/Alpha Vantage · FOMC text/GDELT · Etherscan/CoinGecko/DeFiLlama · ECB/BIS/IMF/World Bank/OECD. **~33 cards × ~820 w avg ≈ Appendix C subtotal: ~27,000 words** ## Appendix D — Style Guide for Empirical-Finance Writing - **D.1** Table craft: layout, what goes in notes, reporting t-stats/SEs/stars, significant digits. `[~3,200 w]` - **D.2** Reporting regressions: which coefficients, $R^2$/within-$R^2$, FE/cluster disclosure, sample lines. `[~2,800 w]` - **D.3** Robustness sections that persuade; the threats-and-responses table. `[~2,400 w]` - **D.4** The replication packet standard (data, code, seed, README, environment file). `[~2,400 w]` - **D.5** Prose style: hedging vs. overclaiming, causal language discipline, citing primary sources. `[~2,200 w]` **Appendix D subtotal: ~13,000 words** ## Appendix E — Solutions Manual (every problem fully worked) > 8 weeks × 5 problem sets × ~6 problems = ~240 problems, each with a complete worked solution > (derivation, code where relevant, interpretation). Counted separately from the problem statements > above (those counts were statement + solution as a pair); this appendix is the canonical home of the > full worked solutions, organized by week. - Solutions, Weeks 1–2 `[~14,000 w]` - Solutions, Weeks 3–4 `[~15,500 w]` - Solutions, Weeks 5–6 `[~14,000 w]` - Solutions, Weeks 7–8 `[~13,500 w]` **Appendix E subtotal: ~57,000 words** **Appendices total: ~124,700 words** --- # CAPSTONE GALLERY (5 full example papers, 8–12 pp each) > Publication-style worked exemplars students can model their own paper on. Each ~3,000–3,800 words > of paper + ~1,000 words of an annotated "how this paper was built" margin commentary. - **Capstone 1 — Fair Lending on HMDA.** Disparity in mortgage approval/pricing; decomposition + a clean design; ties to Sun & Gao (2019). `[~4,600 w]` - **Capstone 2 — Common Ownership from 13F.** Constructing common-ownership measures and a disclosure/competition outcome (a student-track variant of the anchor paper's earnings-management outcome); ties to Gao, Han, Kim & Pan (2024), *JCF*, 84:102520. `[~4,600 w]` - **Capstone 3 — Innovation from USPTO PatentsView.** Patent-based innovation measure and a firm-outcome event study; ties to KPSS (2017). `[~4,400 w]` - **Capstone 4 — SEC 8-K Text Classification.** Classifying 8-K events with OOS-validated text models and a return reaction study; AI-module methods. `[~4,600 w]` - **Capstone 5 — FRED Macro Event Study.** A monetary/macro-announcement event study with proper inference. `[~4,400 w]` **Capstone Gallery subtotal: ~22,600 words** --- # INSTRUCTOR'S MANUAL (separate document) - **IM-1** Pacing guide: the 8-week clock, daily blocks, how to compress to ASSIP 2026's 8-week full-time calendar (Mon-Fri 9-5). `[~3,200 w]` - **IM-2** Grading rubrics (consolidated): problem sets, assessments, capstone paper, presentation. `[~3,400 w]` - **IM-3** Common student pitfalls, by week, with diagnostic questions and fixes. `[~3,600 w]` - **IM-4** Suggested guest lectures & how to slot them; mentor-session facilitation notes. `[~2,400 w]` - **IM-5** Answer keys to assessments + sample graded work (anchor papers for the rubric). `[~3,400 w]` - **IM-6** Equity/access notes: WRDS seats, compute, accommodations; data-licensing compliance. `[~2,000 w]` **Instructor's Manual subtotal: ~18,000 words** --- # GRAND TOTAL — Summary Table | Section | Est. words | |---|---:| | Front Matter | 11,500 | | Week 1 — Probability, Sampling, Inference | 62,000 | | Week 2 — The OLS Engine | 68,000 | | Week 3 — Causal Inference I (PO, matching, IV) | 70,000 | | Week 4 — Causal Inference II (DiD, RD, SC, shift-share) | 73,000 | | Week 5 — Reading the Frontier I | 58,000 | | Week 6 — Reading the Frontier II + AI Co-Pilot | 62,000 | | Week 7 — Research Project I (question → PAP) | 60,000 | | Week 8 — Research Project II (execution → defense) | 61,000 | | Appendix A — Math Toolkit | 14,500 | | Appendix B — Python + LaTeX Setup | 13,200 | | Appendix C — Data Dictionary (~33 cards) | 27,000 | | Appendix D — Style Guide | 13,000 | | Appendix E — Solutions Manual | 57,000 | | Capstone Gallery (5 papers) | 22,600 | | Instructor's Manual | 18,000 | | **GRAND TOTAL** | **~690,800** | > **Word-budget note.** The line-item estimates above sum to ~690,800 words — above the > **300,000–500,000** target. This TOC is intentionally the *maximal* plan so no deliverable is > dropped. The published target is hit by treating the **per-chapter prose floor (≈4,000 w)** rather > than the ceiling, trimming notebook narration to code-comment level, and consolidating solutions, as > follows. **Authoritative budget to write to:** > > | Section | Target words | > |---|---:| > | Front Matter | 10,000 | > | Weeks 1–8 (8 × ~33,000 avg) | 264,000 | > | Appendices A–D | 55,000 | > | Appendix E (Solutions) | 40,000 | > | Capstone Gallery | 18,000 | > | Instructor's Manual | 13,000 | > | **PUBLISHED GRAND TOTAL** | **~400,000** | > > **The book is written to ~400,000 words**, squarely inside the 300k–500k range. Each week is > budgeted at ~33,000 words: 5 chapters at the ~4,000–5,500 w range (technical Weeks 2–4 at the high > end, reading Weeks 5–6 leaner per-guide), 5 problem sets at ~1,800 w each (statement only; full > solutions consolidated in Appendix E at ~40,000 w total), 5 notebooks at ~1,200 w narration each, > one lab/reading pack (~3,500 w), one mentor session (~2,000 w), one assessment (~1,800 w), and the > opening narrative (~1,300 w). Writing slices MUST hit the **target** column; the maximal column is a > ceiling, not a quota. --- ### Open verification items ([CHECK] consolidated) 1. Exact page ranges for the Week 5 classics (FF92, FF93, JT93, Petersen 2009, BDM 2004) — years confident, pages to confirm. 2. Week 6 modern papers' exact pages/venues (KPSS 2017, Hoberg–Phillips 2016, Loughran–McDonald 2011, Bartlett et al. 2022) and **Bhutta–Hizmo–Ringo** venue/year (Fed FEDS working paper vs. journal). 3. All Gao citations are used **verbatim** from `CONVENTIONS.md §6` / spec §5 and are **not** flagged. 4. The POM supply-chain-risk → bank-loan-cost manuscript (ms POM-Nov-25-OA-1924) is **not** used in this TOC (not in the CV selected list per spec §5); excluded until confirmed.