Research Question
To what extent are large language models reliable in conducting discounted cash flow valuations, and which model demonstrates the highest predictive accuracy and analytical quality across diverse industry sectors?
5
AI Models
5
Industries
25
Valuations
45.6%
Avg Error
Abstract
This study investigates the reliability of five large language models (LLMs) — ChatGPT (GPT-4), Claude, Grok, Gemini, and DeepSeek — in conducting Discounted Cash Flow (DCF) valuations across five companies spanning distinct industry sectors: Coca-Cola (Food & Beverage), Sony (Entertainment & Electronics), Tesla (Electric Vehicles/AI), Hasbro (Toys & Games IP), and Microsoft (Cloud Computing/AI Infrastructure). Each model was provided identical standardised prompts containing only historical financial data, with no access to current market prices or analyst consensus. Outputs were evaluated against actual market prices and scored across four criteria: price accuracy (40%), assumption quality (30%), logical consistency (20%), and transparency of reasoning (10%). The study finds that all models systematically undervalued high-growth or AI-premium companies, with Tesla producing near-universal errors of 91–94%. Grok demonstrated the highest average price accuracy, while Claude demonstrated the most methodological diversity. A critical finding is that confidence calibration was poor across all models — with higher confidence assigned to more inaccurate outputs in several cases. The study concludes that LLMs are moderately reliable for stable, predictable businesses but structurally limited in capturing market premiums for optionality, autonomy, and speculative growth.
Introduction
The rise of large language models has introduced a new class of analytical tool capable of processing financial data and producing complex valuations. Professional equity research — long the domain of trained analysts at investment banks and asset managers — is increasingly within reach of AI systems that can perform DCF analysis, sensitivity testing, and investment thesis generation in seconds. However, the reliability of these outputs has not been systematically benchmarked against real-world market prices across diverse industries.
This study addresses that gap. Rather than performing the analysis itself, the researcher acts as the benchmark designer and judge — sourcing real financial data from SEC filings and primary sources, designing a standardised prompt methodology, and evaluating each model's output against verified market prices. The five companies were selected to represent a spectrum of valuation difficulty: from stable cash generators (Coca-Cola) to deeply contested AI-premium stocks (Tesla).
“To what extent are large language models reliable in conducting discounted cash flow valuations, and which model demonstrates the highest predictive accuracy and analytical quality across diverse industry sectors?”
Hypothesis
The study proceeds from the following hypothesis:
“AI models will demonstrate varying levels of reliability across industries, with more stable/predictable businesses yielding more accurate valuations than high-growth or volatile ones.”
This hypothesis predicts that the architecture of DCF analysis — which anchors valuation in historical cash flows and near-term projections — will systematically disadvantage models when applied to companies whose market price reflects speculative future optionality rather than current fundamentals.
Methodology
3.1 Company Selection
| Industry | Company | Ticker | Rationale |
|---|---|---|---|
| Food & Beverage | The Coca-Cola Company | KO | Stable baseline — predictable cash flows, low complexity |
| Entertainment & Electronics | Sony Group Corporation | SONY | Mixed segments, international, FX complexity |
| Electric Vehicles / AI | Tesla, Inc. | TSLA | Controversial, high-growth optionality, chaos test |
| Toys & Games IP | Hasbro, Inc. | HAS | Under-covered, mid-restructuring, stress test |
| Cloud Computing / AI | Microsoft Corporation | MSFT | Large cap tech, clean financials, AI CapEx surge |
3.2 Standardised Prompt Design
The open-ended prompt design required each model to: (1) Select and justify their own valuation methodology (2) State and justify all assumptions from the provided data (3) Show full workings step by step (4) Provide a Fair Value Per Share estimate (5) State a confidence level from 1–10 with reasoning (6) Identify factors that would cause revision of their estimate. This open-ended design was deliberate. By not prescribing a formula, the study allows each model's own reasoning, methodology selection, and assumption-making to emerge naturally — which is itself a data point.
3.3 Scoring Rubric
| Criterion | Weight | Description |
|---|---|---|
| Price Accuracy | 40% | % deviation of AI price target from verified market price on analysis date |
| Assumption Quality | 30% | Reasonableness of growth rates, margins, WACC relative to historical data |
| Logical Consistency | 20% | Internal coherence of the model — do workings match conclusions? |
| Transparency | 10% | Clarity of reasoning, acknowledgement of uncertainty, explanation of methodology |
3.4 Data Sources
| Company | Primary Source | Secondary Cross-Check |
|---|---|---|
| Coca-Cola | SEC Form 8-K FY2023 & FY2024 | Yahoo Finance, MacroTrends |
| Sony | SEC Form 20-F FY2022, FY2023, FY2024 | MacroTrends (USD conversion) |
| Tesla | SEC Form 10-K FY2024, 10-Q Q1 2026 | MacroTrends, TradingView |
| Hasbro | SEC Form 10-K FY2022, FY2024; Hasbro Earnings Release FY2023 | Yahoo Finance |
| Microsoft | SEC Form 10-K FY2022, FY2023, FY2024 | MacroTrends, Morningstar |
Results
4.1 Summary Price Targets
| Company | Market Price | ChatGPT | Claude | Grok | Gemini | DeepSeek |
|---|---|---|---|---|---|---|
| Coca-Cola (KO) | $78.30 | $28.00 | $36.00 | $60.50 | $38.67 | $24.06 |
| Sony (SONY) | $19.78 | $14.10 | $25.00 | $16.00 | $14.19 | $16.10 |
| Tesla (TSLA) | $391.66 | $27.00 | $23.00 | $31.00 | $31.67 | $30.93 |
| Hasbro (HAS) | $95.17 | $96.00 | $67.00 | $105.00 | $84.42 | $135.78 |
| Microsoft (MSFT) | $410.60 | $238.00 | $300.00 | $620.00 | $259.51 | $229.37 |
4.2 Percentage Error
| Company | ChatGPT | Claude | Grok | Gemini | DeepSeek |
|---|---|---|---|---|---|
| Coca-Cola | −64.2% | −54.0% | −22.7% | −50.6% | −69.3% |
| Sony | −28.7% | +26.4% | −19.1% | −28.3% | −18.6% |
| Tesla | −93.1% | −94.1% | −92.1% | −91.9% | −92.1% |
| Hasbro | +0.9% | −29.6% | +10.3% | −11.3% | +42.7% |
| Microsoft | −42.1% | −27.0% | +51.0% | −36.8% | −44.2% |
| Average Absolute Error | 45.8% | 46.2% | 39.0% | 43.8% | 53.4% |
4.3 Confidence Levels
| Company | ChatGPT | Claude | Grok | Gemini | DeepSeek |
|---|---|---|---|---|---|
| Coca-Cola | 6/10 | 5/10 | 6/10 | 7/10 | 6/10 |
| Sony | 6/10 | 4/10 | 5/10 | 7/10 | 5/10 |
| Tesla | 6/10 | 3/10 | 4/10 | 5/10 | 4/10 |
| Hasbro | 7/10 | 6/10 | 6/10 | 6/10 | 7/10 |
| Microsoft | 7/10 | 7/10 | 7/10 | 8/10 | 8/10 |
4.4 Methodology Used
| Company | ChatGPT | Claude | Grok | Gemini | DeepSeek |
|---|---|---|---|---|---|
| Coca-Cola | DCF | DCF + EV/EBITDA | Two-stage DCF | DCF | Two-stage DCF |
| Sony | DCF | SOTP + DCF | Two-stage DCF | DCF | Two-stage DCF |
| Tesla | DCF | Scenario-weighted DCF | Two-stage DCF | DCF | Two-stage DCF |
| Hasbro | DCF | DCF + EV/EBITDA + EV/FCF | Two-stage DCF | FCFE DCF | Two-stage DCF |
| Microsoft | Two-stage DCF | Two-stage DCF | Two-stage DCF | Two-stage DCF | Two-stage DCF |
Key Findings
Tesla Exposed a Universal Structural Failure Across All Models
Every AI undervalued Tesla by between 91.9% and 94.1%, producing price targets of $23–$31 against a market price of $391.66. This was the most consistent and dramatic finding in the entire dataset. The failure was not a modelling error — it was a structural limitation of data-only DCF analysis in the face of speculative market premiums for AI and autonomous driving.
The underlying data made this failure predictable. EBIT margins fell from 16.8% (FY2022) to 7.2% (FY2024), and FCF dropped from $7.33B to $3.58B — while the market priced Tesla at a $1.47 trillion market cap. No retrospective DCF on those numbers can explain a $391 share price without attributing significant value to future autonomous driving revenue that does not yet exist in the financial statements.
“This is the hardest one yet. Tesla is genuinely one of the most valuation-contested stocks in the market — it's priced as an AI/autonomy company by bulls, and as a deteriorating automaker by bears. The data tells an uncomfortable story either way.”
— Claude, Tesla Analysis
“On the numbers alone, Tesla is worth somewhere around $15-22. The market almost certainly prices it higher than $23, and the difference between my estimate and market price is purely the autonomy premium — which is either the biggest opportunity in tech history or the most expensive lottery ticket ever sold, depending on your priors. The data cannot tell you which.”
— Claude, Tesla Analysis
“This valuation is intentionally conservative and reflects current financial reality, not future narratives.”
— ChatGPT, Tesla Analysis
“This is purely data-driven; actual outcomes depend on execution in unmodeled growth areas.”
— Grok, Tesla Analysis
“The gap between OCF and FCF is widening due to CapEx.”
— DeepSeek, Tesla Analysis
Hasbro Was the Most Accurately Valued Company Overall
Hasbro produced the narrowest error range across all five AI models: +0.9% (ChatGPT), −29.6% (Claude), +10.3% (Grok), −11.3% (Gemini), and +42.7% (DeepSeek). ChatGPT achieved near-perfect accuracy at $96.00 against a market price of $95.17.
“Hasbro looks like a stable, cash-generative, low-risk business transitioning into a higher-quality IP model, and based on that, a fair value around $96/share is justified using conservative assumptions.”
— ChatGPT, Hasbro Analysis
“The noise is explainable. The $1,846M FY2023 impairment is a non-cash accounting event — it doesn't affect FCF. The revenue decline is a deliberate divestiture, not a demand collapse. Once you strip those out, the underlying business tells a consistent story: FCF tripling while the business is actively shrinking is a powerful signal of quality improvement.”
— Claude, Hasbro Analysis
“The company has low and declining debt, improving cash generation, and a low beta, making intrinsic cash flow valuation more reliable than multiples.”
— Grok, Hasbro Analysis
“Terminal value = Terminal FCF / (WACC – g_terminal) = 809.4 / (0.058 – 0.025) = 809.4 / 0.033 = $24,527M”
— DeepSeek, Hasbro Analysis
Grok Demonstrated the Highest Average Price Accuracy
Across all five companies, Grok produced the lowest average absolute error at 39.0%. Its strongest performance was on Coca-Cola (−22.7%), Hasbro (+10.3%), and Sony (−19.1%). However, Grok was also the most extreme outlier on Microsoft, overshooting by +51.0% with a $620 price target. Grok's high variance — best on Coca-Cola, most extreme on Microsoft — characterises it as a high-risk, high-reward model.
“Revenue growth: 2025: 3.5% (between 2024's 2.85% and avg ~4.6%); then tapering to 3.0%, 2.8%, 2.6%, 2.5%. Justification: Historical avg ~4.6% but decelerating sharply; mature non-alcoholic beverages industry implies low-single-digit long-term.”
— Grok, Coca-Cola Analysis
“Years 1-5 (FY25-29): 15% CAGR. FY24 showed ~15.7% growth with acceleration; Azure is the primary driver and CapEx is 'intentional and expected to continue.' This is conservative vs. recent momentum but accounts for potential base-effect slowdown. Terminal growth (post-Year 10): 4% (long-term nominal GDP + productivity gains from AI/software; sustainable given historical trends).”
— Grok, Microsoft Analysis
DeepSeek Was the Most Consistently Pessimistic Model
DeepSeek produced the lowest price targets on three of five companies and the highest average absolute error at 53.4%. Its defining characteristic was consistently using the most recent reported FCF as the base without normalising for one-time items or structural changes.
“Base year FCF: $5,296M (FY2024). Given; steep drop from 2023 ($9,755M). Use most recent.”
— DeepSeek, Coca-Cola Analysis
“FCF trajectory is poor: $7.33B → $4.36B → $3.58B despite flat revenue. If this continues, the DCF overvalues. Base FCF normalisation is highly subjective.”
— DeepSeek, Tesla Analysis
“FCF volatility is severe: FY2023 negative –$2.2B → no clear normalised earnings power. Base FY2024 $4.96B might be a one-off recovery.”
— DeepSeek, Sony Analysis
Claude Was the Only Model to Use Non-Standard Methodologies
Every other model applied a standard DCF framework to all five companies. Claude was the only model to adapt its methodology based on the nature of each business. Despite the methodological sophistication, this did not translate into superior price accuracy — Claude ranked fourth overall at 46.2% average error.
“Sony is not a single-business company. A pure DCF on consolidated numbers would mask the fact that gaming (PlayStation) trades at very different multiples than consumer electronics or financial services. SOTP is the right approach here — value each segment by what the market would pay for a pure-play equivalent, then aggregate.”
— Claude, Sony Analysis
“A single-point DCF is useless for Tesla. The range of plausible outcomes is so wide that picking one base case would be intellectually dishonest. Instead I'll build three distinct scenarios — Bear, Base, Bull — each with its own FCF trajectory and terminal assumptions, then probability-weight them.”
— Claude, Tesla Analysis
“I'll run a DCF on the post-restructuring normalized business, then cross-check with EV/FCF and EV/EBITDA multiples appropriate for a branded consumer IP company.”
— Claude, Hasbro Analysis
“The right answer is to weight heavily toward multiples here. The DCF's low WACC is mechanically derived from a low beta that may not fully price in Hasbro's financial distress risk given its leverage.”
— Claude, Hasbro Analysis
Confidence Calibration Was Inverted Across All Models
One of the most significant findings of this study is that AI confidence scores did not correlate positively with accuracy. In several cases, the relationship was inverted — models expressed higher confidence on companies they valued less accurately. All models expressed their highest confidence (7–8/10) on Microsoft, where they produced their largest absolute dollar errors. This pattern — high confidence on familiar-seeming, data-rich companies regardless of whether the model captures the full valuation story — suggests that AI confidence scores reflect perceived data quality rather than actual proximity to market price.
“Confidence Level: 8/10. Why: Microsoft demonstrates remarkably consistent execution. The 15.7% revenue growth in FY2024 is high-quality and backed by 50%+ EBITDA margins.”
— Gemini, Microsoft Analysis
“Confidence level: 8 / 10. Why high confidence (8): Strong historical FCF growth — even with CapEx tripling over 3 years, FCF rose 14% from FY2022 to FY2024. Revenue acceleration (15.7% in FY2024) driven by Azure/AI — real data, not speculation.”
— DeepSeek, Microsoft Analysis
“Confidence: 3/10. The lowest of the three for good reason.”
— Claude, Tesla Analysis
“Confidence Level: 7/10. Rationale: The company has very predictable revenue streams and a low Beta, which provides high confidence in the WACC and terminal growth assumptions.”
— Gemini, Coca-Cola Analysis
The FCF Normalisation Problem Divided Models on Coca-Cola
Coca-Cola's FY2024 FCF dropped 46% from $9,755M to $5,296M, with CapEx rising from $1,852M to $2,064M — a difference that alone does not explain the ~$4.5B FCF decline. How each model handled this anomaly was one of the clearest points of differentiation in the study. This single assumption decision produced a $14 spread in fair value between DeepSeek ($24.06) and Grok ($60.50).
“FCF dropped significantly: 2023: $9.76B; 2024: $5.30B (-46%). This suggests: Either temporary investment (CapEx rising) or structural margin pressure. I will normalize FCF, not blindly use 2024.”
— ChatGPT, Coca-Cola Analysis
“I'll treat FY2024 as an anomaly and use a normalized FCF approach. I'll use $8,200M as the base FCF — closer to the normalized run-rate, acknowledging FY2024 was likely depressed.”
— Claude, Coca-Cola Analysis
“Base year FCF: $5,296M (FY2024). Given; steep drop from 2023 ($9,755M). Use most recent.”
— DeepSeek, Coca-Cola Analysis
Sony's Complexity Exposed the Widest Intra-Model Variance
Sony produced the widest spread of price targets among the five companies: $14.10 (ChatGPT) to $25.00 (Claude) — a $10.90 range against a market price of $19.78. Claude's SOTP approach — the only non-DCF methodology applied to Sony — produced the closest estimate at +26.4% error.
“I'll weight these: 65% SOTP + 35% DCF. Blended fair value = (0.65 × $30.15) + (0.35 × $14.97) = $19.60 + $5.24 = ~$24.80. I'll call it $25 per share (post-split).”
— Claude, Sony Analysis
“Confidence: 4/10. Lower than KO for several structural reasons: The FX problem is huge. Sony reports in yen. The USD translation uses ¥115 → ¥135 → ¥151 over three years — a 31% yen depreciation. This means in yen terms, Sony's business may actually be growing decently, but the USD financials make it look flat.”
— Claude, Sony Analysis
“FX Volatility: Since Sony reports in Yen but is analyzed here in USD, a significant strengthening of the Yen would increase the USD-denominated value of their cash flows.”
— Gemini, Sony Analysis
Microsoft's CapEx Surge Was Interpreted Very Differently Across Models
Microsoft's FY2024 CapEx of $44.5B — nearly double the prior year's $28.1B — was the central analytical challenge in the Microsoft prompt. Models that treated this as a structural cost produced lower valuations; models that treated it as a front-loaded investment produced higher ones.
“The key tension: CapEx nearly doubled in two years ($23.9B → $44.5B) while FCF actually grew — which means operating cash flow grew fast enough to absorb the CapEx surge and still generate more FCF than before. That's an extraordinary result. The business is genuinely accelerating, not just spending.”
— Claude, Microsoft Analysis
“Microsoft's FY2024 CapEx jumped to $44.5B (vs $28.1B in FY2023). This is an investment, not deterioration. If CapEx were normalised to, say, $30B (pre-AI run-rate), FY2024 FCF would be $118.5B OCF – $30B = $88.5B.”
— DeepSeek, Microsoft Analysis
“CapEx: Starts high at 18% of revenue in Years 1-3 (FY24 was 44.5B / 245.1B ≈ 18.2%; reflects ongoing AI buildout), then declines to 12% in Years 4-7, and 8% in terminal (normalization as infrastructure scales).”
— Grok, Microsoft Analysis
“CapEx-heavy now, but intentionally (AI infrastructure). Less suitable for relative valuation since we're restricted from external comps.”
— ChatGPT, Microsoft Analysis
Overall Accuracy Ranking
| Rank | Model | Avg Abs Error (All 5) | Avg Abs Error (Ex-Tesla) | Best Company | Worst Company |
|---|---|---|---|---|---|
| 1st | Grok | 39.0% | 25.8% | Coca-Cola (−22.7%) | Microsoft (+51.0%) |
| 2nd | Gemini | 43.8% | 31.8% | Hasbro (−11.3%) | Coca-Cola (−50.6%) |
| 3rd | ChatGPT | 45.8% | 34.0% | Hasbro (+0.9%) | Coca-Cola (−64.2%) |
| 4th | Claude | 46.2% | 34.3% | Microsoft (−27.0%) | Tesla (−94.1%) |
| 5th | DeepSeek | 53.4% | 43.7% | Sony (−18.6%) | Coca-Cola (−69.3%) |
Note: The ex-Tesla figures remove Tesla from the calculation on the grounds that all models failed for an identical structural reason — the impossibility of capturing AI/autonomy optionality in a data-only DCF. Tesla's universal failure represents a systemic limitation of the DCF methodology itself rather than a differentiating factor between models.
Answering the Research Question
6.1 Reliability
The answer is: moderately reliable, with significant and predictable limitations. Across 25 valuations (5 models × 5 companies), the average absolute error was 45.6%.
When market price reflects future business lines not in current financials (Tesla: 91–94% error universal), DCF is structurally incapable of capturing value. This is not an AI failure — it is a methodology failure that human analysts also face.
When recent financials are distorted by non-cash impairments or divestitures (Hasbro: $1,846M non-cash charge, eOne divestiture), models that identify and strip the distortion perform significantly better. ChatGPT's near-perfect $96 Hasbro estimate demonstrated this.
Even Coca-Cola — which should be the easiest company to value by DCF — produced errors of 22–69%. The 46% FCF anomaly in FY2024 was enough to destabilise even straightforward valuations for models that did not normalise it.
The hypothesis is confirmed: more stable businesses produced more accurate valuations. But the magnitude of even the "easier" errors suggests LLMs should not be used as standalone valuation tools without human review of key assumptions.
6.2 Which Model is Most Reliable?
By raw price accuracy
Grok (39.0% average absolute error, 25.8% excluding Tesla).
By methodological quality and analytical depth
Claudedemonstrated the most sophisticated adaptation to each company's unique characteristics. However, analytical sophistication did not translate into better price accuracy, suggesting that more complex models are not necessarily more correct.
By confidence calibration
No model performed well. All models expressed their highest confidence (7–8/10) on Microsoft, where they produced their largest absolute dollar errors. Claudewas the most honestly calibrated — giving 3/10 on Tesla and 4/10 on Sony, its two least accurate outputs. DeepSeek's 8/10 confidence on Microsoft against a −44.2% error was the most extreme miscalibration in the study.
By consistency
DeepSeek was the most internally consistent but this consistency came at the cost of performance on normalisation-sensitive companies. Grok was the least consistent, producing both the best result (Coca-Cola) and the worst (Microsoft).
Conclusion
Large language models can perform technically correct DCF calculations. They apply CAPM, build FCF projection tables, calculate terminal values using the Gordon Growth Model, and bridge from enterprise value to equity value per share — all without error in the mechanical sense. In that narrow respect, they are reliable.
Where they are unreliable is in judgement: whether to normalise a one-time FCF drop, whether to apply SOTP to a conglomerate, whether the market is pricing a company on its current financials or its speculative future, and whether their own confidence in an output is warranted. These are the decisions that separate a good analyst from a calculation engine — and the data shows that LLMs vary significantly in how well they make them.
The most important finding for any practitioner is the confidence calibration problem. An AI that says it is 8/10 confident in a valuation that is 44% off the market price is not a useful risk signal. Until confidence scores are better calibrated to actual accuracy, AI-generated valuations carry hidden risk that users may not appreciate.
The most important finding for researchers is the Tesla result. Every model failed by a near-identical margin — not because of a bug in their reasoning, but because the question they were asked (value this company from its financial statements) cannot produce the answer the market has given ($391.66) without incorporating information not contained in any income statement, balance sheet, or cash flow statement. That is a limitation of the framework, not the models. And it is a limitation that any AI-powered financial tool must be honest about.
References
Primary Financial Data Sources
Coca-Cola
- —The Coca-Cola Company. (2024). Form 8-K: Fourth Quarter and Full Year 2024 Earnings Release. U.S. Securities and Exchange Commission.
- —The Coca-Cola Company. (2023). Form 8-K: Fourth Quarter and Full Year 2023 Earnings Release.
- —MacroTrends. (2026). Coca-Cola EBITDA and Revenue Historical Data.
- —Yahoo Finance. (2026). Coca-Cola (KO) Balance Sheet and Cash Flow Statement.
Sony
- —Sony Group Corporation. (2022–2024). Annual Report on Form 20-F for Fiscal Years ended March 31.
- —MacroTrends. (2026). Sony Long-Term Debt Historical Data.
Tesla
- —Tesla, Inc. (2024). Annual Report on Form 10-K for Fiscal Year ended December 31, 2024.
- —Tesla, Inc. (2026). Quarterly Report on Form 10-Q for Quarter ended March 31, 2026.
- —MacroTrends. (2026). Tesla Depreciation and Amortization Historical Data.
Hasbro
- —Hasbro, Inc. (2022, 2024). Annual Report on Form 10-K.
- —Hasbro, Inc. (2024). Q4 and Full Year 2023 Financial Results Press Release.
- —Yahoo Finance. (2026). Hasbro (HAS) Cash Flow Statement and Balance Sheet.
Microsoft
- —Microsoft Corporation. (2022–2024). Annual Report on Form 10-K for Fiscal Years ended June 30.
- —Morningstar. (2026). Microsoft Corporation (MSFT) Stock Quote and Fair Value.
Market Price References
- —Robinhood Markets. (2026, May 5). Microsoft (MSFT) Stock Price.
- —Morningstar. (2026, May 4). Microsoft Corporation (MSFT) — Price $413.62.
- —Capital.com. (2026, May 4). Microsoft Corp (MSFT) Market Cap — $3.07T.
- —TradingView. (2026). Tesla (TSLA) Current Price and Market Data.
Valuation Methodology
- —Damodaran, A. (2024). Equity Risk Premiums (ERP): Determinants, Estimation and Implications. Stern School of Business, NYU.
- —U.S. Department of the Treasury. (2026, May). Daily Treasury Par Yield Curve Rates — 10-Year Yield.
- —Internal Revenue Service. (2024). Corporate Tax Rate — 21% Federal Rate. U.S. Tax Cuts and Jobs Act (2017).