Original assignment (Corporate Finance (FIN 4900 capstone) · Upper-division undergraduate · 75-minute session)

Build a DCF valuation of Tesla and write a 5-page equity research report recommending buy/hold/sell.

AI-Resistant Redesign: DCF Valuation & Equity Research Report

1. Redesigned Assignment Overview

Learning Outcomes: Students will construct a defensible DCF valuation model by gathering real financial data, making transparent assumptions, testing sensitivity, and communicating an actionable investment thesis to institutional peers.

Discipline Norms: Professional analysts publish detailed assumption sheets, disclose data sources (10-K filings, Bloomberg, CapItal IQ), show model logic in exhibits, and defend recommendations under peer scrutiny. Version control and audit trails are standard; recommendations change when material inputs shift.

AI Use Stance: AI is allowed as a brainstorming partner, feedback coach, and error-checker on drafts. The grade rests entirely on what you produce, explain, and defend without AI in the room. Your work must be traceable: we will see your sources, your model logic, and hear you explain every major number.

2. AI-Use Map

Stage	AI Level	AI May Do	You Must Produce
Planning & Research	Full	Model a research protocol; brainstorm key value drivers; suggest how to organize your data sources	A completed [teacher-provided research log template] naming ≥5 sources (e.g., Tesla 10-K, investor presentations, industry reports) with URLs and retrieval dates; written three-sentence thesis before you build the model
V1 Model & Report	Shared	Sanity-check formulas; flag circular logic; ask "why that number?" via Socratic prompts	A live, auditable Excel/Google Sheets model with cell references, assumption input table, and five-page report with data annotations and model exhibits
V1 Feedback & Revision	Shared	Offer revision suggestions; quiz you on weak points; help you articulate the delta between V1 and V2	A marked-up [research log update] and a V2 memo (1 page max) documenting what changed, why, and how it affects your recommendation
Final Model Review (in class)	Off	None. You present your final model live, answer peer and instructor questions, and show the model on screen.	A timed, proctored model walk-through (10 min) with no AI notes visible; you field live questions on any input, formula, or assumption
Oral Defense	Off	None. You respond to instructor questions drawn from your numbers and logic.	5–7 min recorded or live response to three custom questions (template-driven, using your actual assumptions and outputs)

3. In-Class Key Work (75 Minutes)

Class Agenda:

Cold Check (10 min, ungraded): Before any review, students answer three items live in a form (not shown):

"What is the most common error in a two-stage DCF when terminal growth rate is set equal to GDP?"

"Name one reason Tesla's WACC might differ from the S&P 500 average."

"If EBITDA grows 2% less than your base case, walk through which line in your model changes first."

Instructor scans responses; don't review answers or show correct key.

Model Presentation & Live Walkthrough (40 min):

Students present their Excel model on the projector in real time (no prerecorded video, no slides hiding formulas).

Instructor asks one live follow-up per student: "Walk me through cell C15 and why that assumption holds."

Peer students note one question each (written on index card, collected).

Peer Pressure Test (15 min):

Instructor reads aloud three peer questions per student (randomized, see card pool).

Student answers without reopening slides or checking notes—they own the logic.

Instructor documents whether the answer shifts the recommendation and by how much.

AI-Off Checkpoint (10 min):

Students complete a timed, unseen-before calculation scenario on paper (or screen, but AI tools disabled):

"If operating margins compress by 200 basis points and terminal growth falls to 1.5%, recalculate your range of equity values. Does your recommendation change? At what margin does it flip?"

No model reference; must show reasoning.

4. Graded Process Checkpoints

Checkpoint	Submit Where	AI Level	What You Submit	Weight %
Research Log (due before V1)	Canvas	Off	Named sources (≥5), thesis statement (3 sent.), question bank for your assumptions	5
V1 Model & Report	Canvas (Excel + PDF)	Shared	Live model, 5-page report, assumption table, sensitivity table (3×3 base/bull/bear on WACC and growth rate)	20
V1 Feedback	Email or LMS chat	Shared	Instructor summary of strengths/blind spots (1 para); peer index cards (posted to Canvas)	—
V2 Research Log & Revision Memo	Canvas	Off	Updated log with revisions explained and cited; 1-page memo: "We changed [X] because [reason]. Impact: [new range for equity value and recommendation if it shifted]."	15
Class Model Presentation & Peer Challenge (live, in class)	Observed in session	Off	10 min model walk; answer one instructor Q + three peer Qs live without notes	20
AI-Off Checkpoint (in class, timed, proctored)	Collected in class	Off	Recalculated scenario range, recommendation if changed, threshold for flip	15
Oral Defense (recorded or scheduled 5 min call post-class)	Canvas or Zoom	Off	5–7 min response to three custom Qs about your numbers and drivers	25

Total: 100% (revision and checkpoint carry 55%).

5. Oral Defense

Pre-Class Preparation:

After the AI-off checkpoint, instructor fills in question templates from the student's V2 submission:

Question Templates (Instructor Fills Brackets):

"Your base-case equity value per share is [student's number], with a bull-case high of [X] and bear-case low of [Y]. Walk me through the three biggest drivers of the range. Which one, if it moved 10 percentage points, would flip your recommendation from [their rec.] to the opposite?"

"You assumed [input, e.g., 'free cash flow growth of 8% for years 2–5']. Take me back to where that came from. What does the 10-K say, and how did you bridge your number to it?"

"Your model shows WACC at [X]%. If Tesla's equity beta moved from [V1 beta] to [higher/lower] because of [market event or sector shift], recalculate the impact on your recommendation. Is it still a [their rec.], or does it flip?"

Format Options (Choose One Per Class Size):

Small class (≤20): Individual 5-min recorded video submitted to Canvas; instructor watches and scores on rubric within 48 hours.

Medium class (20–40): Triples (3 students per defense session) spread over two post-class office hours (30 min each triple = ~6 min per student). Each student answers one custom question; peers listen and note one strength and one clarification needed (anon. feedback form, collected, shared with student and instructor).

Large class (>40): TA-led recorded defenses (TA uses same question templates and rubric); TA scores tier-1 (top and bottom ~20% by V2 score), instructor spot-checks TA scoring and conducts live defenses for borderline cases (10–15 students).

6. Rubric

Criterion	4: Mastery	3: Proficient	1: Needs Work
Disciplinary Reasoning (DCF Logic & Transparency)	Model is fully auditable: cell formulas are labeled, assumption table is separate, two-stage logic is explicit and justified by industry/firm data. Sources are cited in-line or in footnotes. WACC derivation shows beta, cost of equity, and WACC calc with clear inputs. Example: "Beta = 1.8 per Yahoo Finance (retrieved 2/15); Rf = 2.1% (10Y UST); Rm = 7% (historical S&P 500). WACC = 0.6 × 9.5% + 0.4 × 4.2% = 7.9%."	Model is clear and mostly auditable; assumption sources are named but not always cited in full. WACC derivation is shown but one input (e.g., beta or cost of debt) is assumed without backup. Two-stage logic is defensible but terminal growth rate justification is thin (e.g., "1.5%, in line with GDP growth" but no hedge for cyclicality).	Model lacks transparency: formulas are hard to follow, assumptions are embedded and not listed separately. Sources are absent or vague (e.g., "industry report"). WACC is stated but not derived, or uses unjustified numbers. Terminal value calculation is asserted without justification.
Evidence & Data Synthesis	Report weaves ≥5 credible sources (e.g., latest 10-K, investor day presentation, recent analyst report, industry benchmarks, market data). Each major assumption (revenue growth, margins, capex/sales, tax rate) is anchored to company data or peer median. Sensitivity table shows how recommendation holds or shifts across realistic scenarios (e.g., bull: margins improve 100 bps + WACC down 50 bps; bear: growth cuts in half + WACC up 150 bps).	Report cites ≥4 sources, but one or two are generic or dated (>6 months old, not material updates). Most assumptions are sourced, but one (e.g., capex or tax rate) is assumed without explicit backup. Sensitivity table is present but scenarios are narrow (e.g., only ±1 WACC) or assumption changes are not justified as realistic.	Report cites <4 sources or sources are not credible (e.g., stock message boards, unnamed blogs). Assumptions are not sourced; appear arbitrary. No sensitivity table or only trivial one (e.g., ±5% across the board). Scenarios are not realistic or lack a basis in the firm's actual risk profile.
Growth from V1 to V2 (Revision & Closure)	V1 → V2 memo is explicit: "We revised [assumption] because [instructor feedback or new data], and equity value moved from $[X] to $[Y], recommendation remains [rec.] / shifts to [new rec.]**." Changes are substantive (not cosmetic): new source, recalculated input, or logic correction; at least one change is material (>5% impact on value). Research log is updated with retrieval dates and changes noted.	V1 → V2 memo explains 1–2 changes; at least one is material (3–5% impact). Justification is sound but brief. Research log is updated but not all changes are clear. Recommendation is stable or changed, and the memo addresses why.	V1 → V2 memo is missing or vague ("We tweaked assumptions"). Changes are minor or unjustified. Research log is not meaningfully updated. Student cannot explain in the oral defense what changed or why.
AI-Off Performance (Model Walk & Timed Scenario)	In-class model walk: student navigates model without notes, explains formulas and assumptions fluently, answers three peer questions correctly on the spot. Timed checkpoint: recalculated scenario is correct or nearly so; reasoning is shown; threshold for recommendation flip is identified and realistic. No contradictions between model and live explanation.	In-class walk: student explains most formulas; one assumption is misremembered but corrected when asked. Answers 2 of 3 peer questions correctly. Checkpoint: recalculation is mostly right; reasoning is shown but one step is unclear or missing; threshold is approximate but in the right ballpark.	In-class walk: student struggles to navigate the model, reads numbers but doesn't explain logic, answers <1 of 3 peer Qs correctly. Checkpoint: calculation is incomplete, incorrect, or reasoning is absent; threshold is not identified. Student asks for notes or model during the walk.
Oral Explanation & Defense	Student answers three custom Qs fluently: traces each number back to a source or model cell, explains how scenarios shift the recommendation, identifies a specific threshold (e.g., "If WACC > 8.5%, recommendation flips to hold"). Acknowledges a limitation (e.g., "We assume margin stability, but cyclical downturns could compress them; that's a key risk."). No hesitation or contradictions with earlier claims.	Student answers all three Qs and cites the right numbers and logic; one explanation is slightly unclear or needs a follow-up Q to clarify. Acknowledges at least one assumption limitation. Minor gap between oral and written work.	Student answers 1–2 of 3 Qs; numbers are uncertain or inconsistent with the model. Cannot trace an assumption to its source. No acknowledgment of risk or assumptions. Oral explanation contradicts the report or model.
Presentation & Clarity	Report is concise (5 pages), well-organized (executive summary → thesis → analysis → exhibits → recommendation). Figures are labeled, models are readable. Writing is clear and direct; financial language is precise. No typos or formatting errors. Exhibits (DCF table, sensitivity grid, peer comps if included) are professional and cited.	Report is 5–6 pages, organized clearly, but one section is dense or exhibit is hard to read. Writing is mostly clear; minor imprecisions in language (e.g., "free cash flow" used loosely). 1–2 typos. Exhibits are present and mostly professional.	Report is >6 pages or poorly organized (hard to follow the logic). Writing is unclear or uses imprecise financial language. Multiple typos. Exhibits are missing, hard to read, or not labeled.

Automatic Caps:

No recommendation (buy/hold/sell) or no clear thesis: All criteria capped at 2.

No evidence (sources, model backup, sensitivity): "Evidence & Data Synthesis" capped at 1; all other criteria capped at 2.

Model is not submitted or not auditable (formulas hidden, no assumption table): "Disciplinary Reasoning" and "AI-Off Performance" capped at 1.

Did not complete V1 → V2 revision or checkpoint: "Growth" and "AI-Off Performance" capped at 1.

Report >7 pages or <4 pages: "Presentation" capped at 2.

7. Syllabus & AI-Use Statement

Add to Syllabus / Canvas Course Policy:

Artificial intelligence (AI) tools are permitted in this capstone to help you learn, as described below. Using AI this way is never cheating.

AI is your brainstorming partner and feedback coach: You may use generative AI (e.g., ChatGPT, Claude) to help you organize a research plan, receive Socratic feedback on draft assumptions, or check Excel formulas for errors. You must disclose which AI tools you used in a one-line note on the coversheet of each submission (e.g., "Brainstormed research sources with ChatGPT; checked formulas with Google Sheets AI; received editorial feedback on V1 report from Claude").

AI is not allowed in the graded checkpoints: Your in-class model presentation, the AI-off timed scenario, and your oral defense are solo work. You must produce and explain every number and formula without AI assistance or notes during these events.

Your grade depends on what you can show, explain, and defend without AI. Revisions that respond to feedback carry the most weight (15% + 25% across the rubric). The AI-off checkpoint and oral defense (40% combined) are evaluated separately and must demonstrate mastery of your own logic.

Violations of this policy: Submitting AI-generated text as your own (e.g., a report paragraph copied from a chatbot without disclosure) is academic dishonesty under the [Institution] Academic Integrity Policy. So is using AI during a checkpoint or defense when it is forbidden, or fabricating AI use to excuse weak work.

Questions? See me in office hours or post in the Canvas discussion board.

8. Why This Resists AI

Research log + source tracing: AI cannot fabricate credible 10-Ks, Bloomberg pages, or investor presentations. Students must name real documents and show retrieval dates; cross-checking is trivial.

Assumption transparency + audit trail: Every formula is visible and linked to real data. AI-generated numbers (which are often plausible-sounding fiction) are caught the moment a student is asked to cite the source.

Scenario sensitivity with threshold reasoning: Writing "If WACC changes by X, recommendation flips at Y" requires deep understanding. AI can generate a sensitivity table, but reasoning about which threshold matters and why is student-owned and defended orally.

Live model walkthrough (no notes, on screen): AI cannot impersonate the student or answer live questions in real time. This 10-minute walk is the most forensic part of the assessment.

AI-off checkpoint (timed, proctored, scenario-based): This forces the student to recalculate under a new constraint without AI. No amount of prompt engineering recovers a weak understanding of the DCF logic.

Oral defense with custom Qs from the student's own numbers: Questions are not generic; they reference the exact assumptions, formulas, and outputs in the student's model. AI cannot pre-answer these because they do not exist until the student's V2 is submitted.

Grading weight on revision + live components: 55% of the grade is earned through V1→V2 improvement plus live/proctored checkpoints. Even if V1 is AI-assisted, the delta and the defense carry the points.

Peer pressure + index cards: Peer questions are random and written by classmates who have seen the model. Surprise questions reward genuine mastery, not memorized talking points.

Practical Workload Notes:

For ≤25 students: Conduct all oral defenses live in-class or office hours (5–6 hours total instruction time; TA can co-observe).

For 25–60 students: Use recorded video defenses (students post 5 min video, instructor grades on rubric in 3–5 min per student; ~30–60 min total grading).

For >60 students: TA scores all videos using the rubric; instructor spot-checks bottom and top 20% and attends live re-dos if needed for borderline cases (reduces oral load to ~8–10 defenses).

Redesign your own assignment in about a minute. 3 free per month.
Try the redesigner