A Fundamental and Bayesian Framework for Equity Valuation and Investment Decisions
Author
Bat-Erdene | 114035128
Published
November 20, 2026
Abstract
This project develops a transparent method for investigating equity investment opportunities using financial statement analysis, dividend valuation, and Bayesian learning. The central question is whether incorporating uncertainty about dividend growth produces more informative investment assessments than relying on a single historical estimate. The project distinguishes the return required by an investor from the return implied by a market price and a forecast dividend stream. A reproducible research design is proposed for dividend-paying firms, with dated information, explicit sampling rules, forecast evaluation, and investment performance tests after transaction costs. A completed hypothetical case illustrates the calculations. With the selected inputs, Bayesian updating gives a mean implied-return estimate of 9.50%, with a 95% credible interval from 5.84% to 13.15%. The probability of exceeding the selected hurdle is 60.51%. No real-company dataset or historical investment backtest is claimed. The contribution is a practical methodology linking valuation assumptions, statistical uncertainty, and testable investment decisions.
Keywords: investment research; equity valuation; dividend discount model; implied cost of equity; Bayesian inference; research methodology.
Project status: Completed methodological study and numerical illustration, with a proposed empirical validation protocol. Inspired by the structure and cost-of-capital theme of the supplied draft [1].
Reproducibility note: All example results, tables, and charts are calculated within this file. The default inputs reproduce the original project. Fixed stress scenarios are explicitly identified. Edit the student details in the YAML header before submission. HTML is the default output; Word and PDF are optional.
1. Introduction and Research Problem
Investment research involves translating information about a business into a defensible assessment of value, risk, and possible return. A market price is observable, but future dividends and growth are uncertain. Two analysts can therefore examine the same company and reach different conclusions because their forecasts, required returns, or confidence in available evidence differ.
The supplied paper uses dividend valuation and Bayesian cost-of-capital analysis to connect forecasts with investment decisions [1]. This project adopts that general theme while emphasizing research design and an accessible worked example. It does not reproduce the draft’s theorems, claim new mathematical discoveries, or treat the investment anecdotes in that manuscript as evidence of a generally successful strategy.
1.1 Problem statement
A point estimate can conceal substantial uncertainty. For example, an estimated implied return of 10% may appear attractive relative to a 9% hurdle, yet the estimate may depend heavily on uncertain perpetual growth. A useful research method must show which assumptions drive the result, how uncertainty is measured, and whether the resulting forecasts and decisions perform adequately on unseen observations.
1.2 Aim and objectives
The aim is to design and demonstrate a reproducible framework for evaluating dividend-paying equities. The objectives are to: (1) connect business fundamentals to dividend forecasts; (2) estimate implied returns using a conventional model and a Bayesian alternative; (3) quantify sensitivity to growth and market price; (4) specify tests of forecast accuracy and investment performance; and (5) identify the conditions under which the model should be rejected or replaced.
1.3 Research questions
RQ1: How does combining prior information with observed dividend growth change the estimated implied return and its uncertainty?
RQ2: Does the proposed Bayesian forecast reduce out-of-sample dividend forecast error relative to a historical-growth baseline?
RQ3: Do investment signals based on posterior probabilities add value after costs and risk adjustment?
RQ4: How sensitive are conclusions to dividend cuts, alternative priors, and growth assumptions?
1.4 Scope and contribution
The framework targets common shares with positive, reasonably interpretable cash dividends. The primary contribution is a documented research process, not a claim that Bayesian methods automatically outperform other approaches. The numerical case answers RQ1 and illustrates RQ4. RQ2 and RQ3 require the empirical study described later.
2. Literature and Conceptual Framework
2.1 Dividend valuation
Dividend discount models relate equity value to expected distributions to shareholders. CFA Institute [2] describes constant-growth and multistage approaches and emphasizes the sensitivity of valuation to growth and required return. Dividend models are most useful when payouts have a meaningful relationship to the firm’s economics. This motivates the project’s eligibility screen and its use of alternative models for firms whose dividends are unrepresentative.
2.2 Implied cost of capital
Pastor, Sinha, and Swaminathan [3] investigate the implied cost of capital as a measure relevant to time-varying expected stock returns. The methodological distinction is important: an implied return is obtained by solving a valuation equation given a market price and forecasts. It is not an observed future return. Forecast error or a misspecified terminal growth assumption can make the resulting estimate misleading.
2.3 Bayesian learning
Bayesian inference updates a prior distribution using the likelihood of observed data. In a normal-mean setting, the posterior combines prior information and sample information according to their precision [4]. The framework makes assumptions about prior beliefs explicit. It also permits probability statements about an unknown parameter conditional on those assumptions. Greater apparent precision is useful only when the prior and likelihood are credible.
2.4 Financial information and evidence quality
The SEC’s investor guidance describes annual filings as sources of business information, risk disclosures, management discussion, and financial statements [5]. For this project, such records support the economic plausibility of the dividend forecast. A dividend trend should be interpreted alongside earnings, cash generation, debt obligations, and management’s distribution policy.
2.5 Research gap addressed by this project
The project addresses a practical implementation gap: a valuation formula alone does not specify how to select firms, avoid future information, measure predictive performance, or translate uncertainty into a consistent decision rule. This is a teaching and implementation contribution, not a claim that these topics are absent from the academic literature.
Research stage
Question answered
Project output
Fundamental analysis
Can the forecast be economically justified?
Dated business and payout assessment
Statistical estimation
What growth values remain plausible?
Posterior distribution and sensitivity checks
Valuation
What return does the price imply?
Implied return distribution
Validation
Does the method work on unseen data?
Forecast and portfolio evaluation protocol
3. Valuation Model and Assumptions
3.1 Notation and measurement
All rates in the main example are annual and expressed as decimals in calculations. P denotes the current share price, \(D_{0}\) the most recent full-year dividend per share, g constant expected annual dividend growth, k an investor’s required annual return, and r the annual return implied by price and forecast payouts. Price and dividends use the same currency and share basis.
Equation (2) values the modeled dividend stream at a chosen required return. Equation (3) solves the same relationship for the implied return at the observed price. These established relationships are summarized in CFA Institute [2]. The distinction between y and the forward dividend yield matters: \(D_{1}\)/P equals y(1 + g), not y.
3.3 Maintained assumptions
The illustration assumes P > 0, \(D_{0}\) > 0, and g > -1. Dividends are paid at year-end; the modeled firm continues indefinitely; and one growth rate summarizes its long-run payout path. Taxes, currency effects, trading frictions, and changes in payout policy are omitted from the valuation illustration. They must be addressed separately in an empirical investment test.
With a positive dividend and g > -1, equation (3) gives r - g = y(1 + g) > 0. The price-implied dividend series therefore converges at its solved rate. By contrast, valuing the share at an independently selected hurdle k requires checking k > g. A value computed with k less than or equal to g is not a valid finite Gordon-model valuation.
3.4 Economic interpretation
For fixed expected payouts, a lower price mechanically produces a higher implied return. It does not establish that the investment is safer, nor that payouts will actually be delivered. A price decline may accompany deteriorating fundamentals. The research process must update both the observed price and the forecast assumptions before interpreting a change in attractiveness.
For a company with temporary rapid growth, a multistage model should replace perpetual extrapolation. Firms that do not pay meaningful dividends require a different valuation basis; they are outside this project’s primary sample.
4. Bayesian Estimation and Decision Rule
4.1 Prior and likelihood
Let \(x_{1}\), …, \(x_{n}\) be annual dividend growth observations. As a tractable teaching approximation, assume \(x_{i}\) conditional on g follows a normal distribution with mean g and known variance \(sigma^{2}\). Let the prior for g be normal with mean \(mu_{0}\) and variance \(tau_{0}^{2}\). This conjugate normal model gives the following update [4]:
The prior must be set using information available before the evaluation observations. Reusing the same observations to construct both a prior and its likelihood overstates information. In the illustration, the normal probability below g = -100% is negligible. For general applications, use a properly truncated model or model log dividend growth; the unrestricted normal approximation should not assign material probability to impossible dividend paths.
The approximate 95% credible interval is the posterior mean plus or minus 1.96 posterior standard deviations. This describes uncertainty about the model-implied rate, conditional on the model. It is not a prediction interval for the return an investor will realize. Future price changes, dividend shocks, and model error require additional modeling.
For the demonstration, designate a stock as eligible for further review when p is at least 0.80. The threshold is an illustrative research choice and must be set before testing. A binary loss model can motivate it: if a false positive costs \(C_{FP}\) and a false negative costs \(C_{FN}\), the threshold is \(C_{FP}\)/(\(C_{FP}\) + \(C_{FN}\)). A value of 0.80 therefore corresponds to a four-to-one loss ratio.
This simplified classification rule does not optimize portfolio wealth, position size, or investor utility. A full investment decision also requires payoff magnitudes, dependence across holdings, liquidity, and downside constraints. Passing the screen is an analytical signal, not a guarantee of profit.
5. Research Methodology and Data Plan
5.1 Design and sampling
The proposed empirical extension is a quantitative, longitudinal study using secondary financial data. Select one clearly defined market before collection. Target at least 30 dividend-paying firms across three or more sectors, with at least six consecutive annual dividend observations available at each initial estimation date. Six dividend levels generate five annual growth observations. Eligibility must be reassessed using only the information available at each formation date.
Use 2014-2019 as the initial development period, 2020-2022 for validation, and 2023-2025 as a locked test period. This is a proposed schedule, not a claim that observations have been collected. Apply a fixed June year-end formation rule and include only disclosures published by the formation date. Delay any unavailable inputs; never assume that a fiscal-year observation was known at the fiscal-year end.
5.2 Required variables
Variable
Definition / role
Proposed source
P; dividends
Unadjusted price and split-consistent cash dividends per share
Return series with dividends and corporate actions
Documented market-data source
Benchmark; cash
Market total return and aligned cash return
Index provider; official rate source
5.3 Cleaning and controls
Store each observation with its source URL, release date, currency, unit, and extraction note. Reconcile share splits consistently across price and dividends. Separate ordinary dividends from special distributions, using ordinary dividends for the primary growth estimate and both for realized investor returns. Do not treat missing dividends as zero without confirmation.
Record entry, exit, suspension, and delisting events. Do not select only firms that survive to the end of the sample. If historical constituents or delisting returns cannot be obtained, disclose the resulting coverage limitation. Keep quarterly figures separate from annual totals and do not add cash dividends to an already dividend-adjusted total-return series.
5.4 Ethics and reproducibility
Use public or appropriately licensed data, retain source records, and avoid personal brokerage identifiers. Archive raw data separately from cleaned data. Record all changes, exclusions, assumptions, and model versions so another researcher can reconstruct the analysis.
6. Empirical Tests and Implementation
6.1 Estimation workflow
At each formation date, freeze the available information. Review business fundamentals and confirm model eligibility. Estimate a historical-growth baseline using the available trailing window. Set Bayesian prior inputs from earlier development information, fit the posterior, and compute implied-return summaries. Store the forecast before observing the next dividend or subsequent investment return.
The known-variance model in Section 4 is used only for the demonstration. In the empirical extension, uncertainty about the growth variance should be modeled jointly or assessed by a documented resampling procedure. When prior inputs are estimated from development data, preserve that estimation rule through the locked test and report the additional uncertainty it creates.
6.2 Hypotheses and primary measures
Hypothesis
Testable comparison
Evaluation
H1: Forecast accuracy
Bayesian vs. historical-growth forecast
Lower mean absolute next-year dividend forecast error
H2: Calibration
Nominal vs. observed predictive coverage
Coverage and width of 80% and 95% predictive intervals
H3: Investment usefulness
Signal portfolio vs. eligible-universe portfolio
Net excess return, volatility, drawdown and turnover
For H1, use paired forecast errors on the same firm-dates, reporting raw and scale-normalized errors. For H2, generate predictive intervals that include observation noise; a credible interval for mean growth is not sufficient. For H3, compare the two strategies on matching dates and disclose whether apparent benefits are attributable to sector concentration or risk exposure.
6.3 Portfolio protocol
Rebalance annually after the information cutoff, using next-session execution prices where available. Set k to the contemporaneous cash rate plus a predeclared 5 percentage-point premium, and hold the 0.80 probability threshold fixed. Equal-weight eligible signals with a maximum 10% weight per stock; hold residual funds in cash when fewer than ten names qualify. Use no leverage or short selling. Maintain both a broad-market benchmark and an equal-weight eligible-universe benchmark.
Apply 25 basis points to the value of each purchase and sale, including the initial allocation, and report sensitivity at 10 and 50 basis points. These are assumed cost scenarios, not observed broker charges. Treat delistings explicitly and report results before personal taxes.
6.4 Statistical interpretation
Report effect sizes and uncertainty, not only whether a p-value crosses a threshold. Use date-block resampling with all firms retained within each sampled block to respect common market shocks, and discuss residual dependence. A short annual test provides limited evidence: do not treat many correlated firms as many independent market histories. Strong performance claims require a longer evaluation period or independent replication.
7. Completed Hypothetical Investment Case
All figures in this section are invented for demonstration. Alpha Manufacturing is a fictional mature dividend-paying company. Monetary values are expressed in generic currency units. This is not a company recommendation or an empirical result.
The 5 observations have an arithmetic mean growth rate of 6.00%. The known observation standard deviation is an explicit teaching assumption; it is not the fitted sample standard deviation.
Code
yield_trailing <- params$dividend / params$pricehistorical_implied_return <- yield_trailing + (1+ yield_trailing) *mean(x)knitr::kable(data.frame(Measure =c("Trailing dividend yield", "Historical mean growth","Historical-growth implied return", "Plug-in value at historical growth"),Value =c(pct(yield_trailing), pct(mean(x)),pct(historical_implied_return),if (is.na(baseline_value)) "Not finite under the model"elsenum(baseline_value))), caption ="Historical-growth baseline.")
Historical-growth baseline.
Measure
Value
Trailing dividend yield
4.00%
Historical mean growth
6.00%
Historical-growth implied return
10.24%
Plug-in value at historical growth
141.33
The historical-growth estimate is 10.24%. The plug-in value assumes that historical average growth continues indefinitely. It is not an independently verified intrinsic value. If the growth assumption is at or above the required return, the finite Gordon valuation is not defined.
7.2 Bayesian calculation
The calculation follows equations (4) through (8). The code below exposes the same update used throughout the document.
Prior information receives 35.71% of the weight, while the sample mean receives 64.29%. The posterior mean implied return is 9.50% and its standard deviation is 1.86 percentage points.
The 95% credible interval is 5.84% to 13.15%. The interval crosses the required-return hurdle. This measures uncertainty about the model-implied rate conditional on the assumptions, not the distribution of an investor’s realized future return.
Posterior density of the model-implied annual return. Shading marks returns above the selected hurdle; the orange dashed line is the hurdle.
7.3 Decision interpretation
The probability that the implied return exceeds 9.00% is 60.51%. At a threshold of 80.00%, the fictional stock does not pass the probability screen. This classification is a research screen, not a complete portfolio decision.
The plug-in value at posterior mean growth is 113.38. This is \(V(k,E[g])\), not\(E[V(k,g)]\). Valuation is nonlinear, and the unrestricted normal growth posterior assigns probability to \(g\geq k\), where a finite Gordon valuation at the selected hurdle does not exist. Therefore, this project does not report a posterior mean intrinsic value under that model.
8. Sensitivity and Stress Analysis
8.1 Price and growth scenarios
The following fixed scenario grid uses prices of 80, 100, and 120 and growth rates of 2%, 4%, and 6%. It uses the selected dividend of 4.00. These are deterministic assumptions, not forecasts.
Deterministic implied-return sensitivity to share price at three constant growth assumptions.
At 4% growth, a price decline from 100 to 80 raises the implied return from 8.16% to 9.20% if dividends remain unchanged. However, if the dividend falls by 25% to 3.00 and growth falls to 2%, the implied return at price 80 is 5.82%. A lower price does not by itself establish improved value when expected payouts deteriorate.
8.2 Prior sensitivity
Only the prior mean changes in this table. Observations, price, dividend, prior standard deviation, and observation standard deviation remain at their selected values.
Code
prior_means <-c(0.02, 0.04, 0.06)prior_fits <-lapply(prior_means, function(m) {normal_update(x, m, params$prior_sd, params$observation_sd, params$price, params$dividend, params$required_return, params$credible_level)})prior_table <-data.frame(`Prior mean growth`=pct(prior_means),`Posterior growth`=pct(vapply(prior_fits, function(z) z$growth_mean, numeric(1))),`Mean implied return`=pct(vapply(prior_fits, function(z) z$return_mean, numeric(1))),`Probability above hurdle`=pct(vapply(prior_fits, function(z) z$probability, numeric(1))),check.names =FALSE)knitr::kable(prior_table, caption ="Sensitivity to the prior mean growth assumption.")
Sensitivity to the prior mean growth assumption.
Prior mean growth
Posterior growth
Mean implied return
Probability above hurdle
2.00%
4.57%
8.75%
44.76%
4.00%
5.29%
9.50%
60.51%
6.00%
6.00%
10.24%
74.70%
More optimistic priors raise the implied-return assessment. This is a model response to an assumption, not independent evidence that optimism is justified. Report the sensitivity rather than selecting the prior that produces a preferred investment conclusion.
8.3 Required robustness checks
Repeat estimation with weaker and stronger priors, alternative trailing windows, and separate treatment of special dividends. Check whether unusually large growth observations dominate results. Evaluate a dividend-cut scenario and a multistage payout model. Inspect predictive residuals and structural changes rather than assuming that a long history represents one stable regime.
8.4 What these calculations establish
The illustration establishes the arithmetic relationship between assumptions and model outputs. It does not establish forecasting accuracy, an optimal screening threshold, or superior realized returns. Those questions belong to the proposed locked empirical evaluation.
9. Discussion, Limitations and Conclusion
9.1 Discussion
The completed example illustrates a meaningful difference between an attractive point estimate and a sufficiently confident screening result. The historical-growth implied return is 10.24%, compared with a posterior mean of 9.50%. The posterior interval remains an essential part of the assessment. This supports reporting uncertainty alongside valuation instead of treating a single output as a precise fact.
The method also separates three tasks that are often confused: estimating a firm’s dividend path, solving for a price-implied rate, and evaluating actual investment performance. Success in one task does not establish success in the others. An informative dividend forecast can still produce a poor investment signal if the market price, risk exposure, or valuation model is poorly understood.
9.2 Limitations
Model risk. Constant perpetual growth is restrictive. The method may misrepresent cyclical firms, changing payout policies, or businesses that return capital primarily through buybacks.
Statistical risk. Independent normal growth with known variance is a teaching simplification. Dividend cuts, zero payouts, serial dependence, and structural breaks require richer models.
Prior risk. An overconfident incorrect prior may narrow uncertainty while increasing error. Posterior concentration is not proof of accuracy.
Data risk. Release-date mistakes, missing delistings, share-basis inconsistencies, and selective samples can create false performance.
External validity. A small sample or short test period cannot establish that results generalize across markets and regimes.
9.3 Implementation schedule
Stage
Work product
Week 1
Finalize question, sample, sources and preregistered rules
Week 2
Collect data; audit release dates and corporate actions
Week 3
Estimate baseline and Bayesian models; validate calculations
Week 4
Run validation; lock decisions before test evaluation
Week 5
Evaluate forecasts, portfolios, costs and sensitivity
Week 6
Write findings, limitations and reproducibility appendix
9.4 Conclusion
Investment financial research should make its evidence, assumptions, and evaluation rules explicit. This project demonstrates how fundamental analysis, dividend valuation, and Bayesian learning can be combined in one reproducible framework. Its worked case shows that a return estimate above a hurdle may coexist with insufficient confidence to pass a predefined screen. The next research step is to implement the proposed data protocol and evaluate the framework on genuinely unseen observations. No claim of market outperformance follows from the hypothetical example alone.
References
[1] Wong, W.-K., and Baljingarav. (n.d.). How to Invest in Stocks by Using the New Theory on the Cost of Capital with Real-Life Examples. Unpublished draft supplied by the user, filename: CoC draft1.pdf. Used as a structural and thematic reference; its empirical claims were not independently validated for this project.
[2] CFA Institute. (2026). Discounted Dividend Valuation. Refresher reading. https://www.cfainstitute.org/insights/professional-learning/refresher-readings/2026/discounted-dividend-valuation
[3] Pastor, L., Sinha, M., and Swaminathan, B. (2006). Estimating the Intertemporal Risk-Return Tradeoff Using the Implied Cost of Capital. NBER Working Paper No. 11941. https://www.nber.org/papers/w11941
[4] DiMaggio, C. (n.d.). Intro. Bayesian statistics teaching notes, section: Single-Parameter Normal Model. Columbia University. https://www.columbia.edu/~cjd11/charles_dimaggio/DIRE/styled-4/styled-11/code-4/
[5] U.S. Securities and Exchange Commission. (n.d.). How to Read a 10-K. Investor.gov. https://www.investor.gov/introduction-investing/getting-started/researching-investments/how-read-10-k
Online sources consulted for the original project: 6 November 2026. Numbered references identify the sources used for the conceptual foundation. Numerical examples, sample design, screening settings, and the proposed implementation schedule are constructed for this project.
Appendix A. Reproduction checklist
Set or edit the inputs in the YAML params block. The defaults reproduce the original hypothetical case.
Compute the posterior variance from equation (4), then its square root.
Compute posterior growth from equation (5).
Set y = \(D_{0}\)/P and apply equations (6) and (7).
Form the interval using the normal quantile for params$credible_level and the posterior standard deviation.
Evaluate the normal tail probability in equation (8).
Compare that probability with params$probability_threshold and document the outcome.
Repeat the calculation for each changed prior or price; retain all other inputs unless the scenario explicitly changes them.
Appendix B. Minimum empirical results table
For each model, report the number of firms and firm-dates; forecast mean absolute error; 80% and 95% predictive coverage; net annualized return; annualized volatility; maximum drawdown; turnover; and benchmark-relative performance. Mark unrun analyses as “not estimated.” Do not substitute the worked hypothetical figures for empirical results.
Appendix C. Session Information
The output below records the R environment used when this file is rendered.
Code
sessionInfo()
R version 4.5.1 (2025-06-13 ucrt)
Platform: x86_64-w64-mingw32/x64
Running under: Windows 11 x64 (build 26200)
Matrix products: default
LAPACK version 3.12.1
locale:
[1] LC_COLLATE=English_United States.utf8
[2] LC_CTYPE=English_United States.utf8
[3] LC_MONETARY=English_United States.utf8
[4] LC_NUMERIC=C
[5] LC_TIME=English_United States.utf8
time zone: Asia/Taipei
tzcode source: internal
attached base packages:
[1] stats graphics grDevices utils datasets methods base
loaded via a namespace (and not attached):
[1] htmlwidgets_1.6.4 compiler_4.5.1 fastmap_1.2.0 cli_3.6.5
[5] tools_4.5.1 htmltools_0.5.8.1 rstudioapi_0.19.0 yaml_2.3.10
[9] rmarkdown_2.29 knitr_1.50 jsonlite_2.0.0 xfun_0.53
[13] digest_0.6.37 rlang_1.1.6 evaluate_1.0.5