Most financial analysis techniques fail when a ratio, a model, or a score is treated as a verdict. A CFA Institute survey of investment professionals found that Excel was used by 90% of respondents for valuation tasks, while 20% also used Python in the same workflow. Spreadsheets still carry the judgment. Code should make that judgment rerunnable.
The techniques analysts rely on are a toolkit, not a sequence you must finish. You match the method to the decision, clean the statements so the comparison is fair, and name the point where the method stops being informative. This roundup covers ten core financial analysis techniques, from ratio analysis and DuPont decomposition through Monte Carlo simulation and the Altman Z-Score. Each entry is a working workflow: when to use it, what the data must support, the pitfalls that waste a review, and how to keep the run auditable.
Finzer's guide to financial analysis is a useful map of statement reading before you lock a model. A high current ratio can mean cash strength or unsold inventory. High return on equity can mean efficient operations or unusually high debt. I treat any single figure as a screen. The comparison that holds up uses common-size statements, multi-year trends, peer benchmarks, and a cash check. A short comparison after the tenth technique helps you match method to decision.
1. Ratio Analysis
A printed pack still hides the comparison. The gauges only help after the definitions are fixed.
Ratio analysis compares lines from the income statement, balance sheet, and cash-flow statement so you can judge liquidity, profitability, solvency, efficiency, and valuation. The current ratio, return on equity, debt-to-equity, inventory turnover, and price-to-earnings mean little until you set them against history, peers, and the firm's own policies. Graham and Dodd's Security Analysis (1934) made that comparison a standard of investment work. It is still the first pass I run.
The SEC's data research collection publishes structured datasets from filings, including numbers from statements and notes. You can calculate a margin for several years, place it beside competitors, and ask whether the move came from price, cost, asset use, or financing.
Say you track current ratios for five regional banks across eight years. A steady slide can show liquidity strain while earnings still look fine. For a retailer, gross margin next to inventory turnover, benchmarked to sector medians from filings, often explains a "cheap" multiple better than the multiple itself.
What has to be true first
Practical rule: A ratio without its denominator and definition can't be audited later.
- Clean the panel: Align fiscal year-ends, flag restatements, and log missing lines before you divide.
- Don't snapshot: Multi-year trends and peer ranges beat one year.
- Show the code: A financial data analysis pass in Plan Mode lets you confirm definitions before Python runs.
2. DuPont Analysis
DuPont analysis splits return on equity into net profit margin, asset turnover, and the equity multiplier. The three-step form is ROE = (net income / sales) × (sales / total assets) × (total assets / equity). A five-step form also separates tax burden, interest burden, and operating margin. Use it when ROE moved and you need to know whether profit, asset use, or borrowing did the work.
Two retailers can post the same ROE and still be different businesses. One is high-margin and slow-turning. The other is thin-margin and fast-turning. A bank whose ROE rises from 9% to 13% may look healthier until the split shows increased debt usage, not margin, did most of the lifting. That is the decision the decomposition is for.
How to attribute the change
Use average assets and equity when the period is a full year. Opening balances flatter turnover after a large acquisition. Attribute the change in ROE with a sequential substitution or a log-based split, and state which method you used. Reviewers argue about the method more often than about the arithmetic.
Pair the split with a leverage sensitivity. If a one-turn drop in the equity multiplier wipes out the ROE gain, you don't have an operating story.
PlotStudio AI can save the decomposition as an Analysis Page with the charts and the Python, and you can export a Jupyter notebook when a colleague needs to rerun the drivers.
3. Discounted Cash Flow (DCF) Valuation
DCF values an asset as the present value of expected free cash flows, discounted at a rate that reflects risk, usually WACC. A terminal value covers cash flows past the explicit horizon. The time-value idea is older than the equity models. Irving Fisher framed it, and John Burr Williams formalized equity valuation in The Theory of Investment Value (1938).
Three fundamental approaches sit beside each other in Stephen Penman's discussion of financial statement analysis and security valuation: the discounted dividend model, the discounted cash-flow model, and the residual-income model. Free cash flow to the firm is cash for all capital providers. Free cash flow to equity is what remains for common shareholders after investment and financing effects. Forecast quality depends on whether margins, working capital, capital spending, and growth are sustainable, not on the discount formula.
A manufacturer with a five-year forecast and a 2% perpetuity growth rate is a standard setup. So is a project valued on incremental free cash flow at a project hurdle rate, not at the firm's blended WACC, when the risk differs.
Report terminal value as a share of enterprise value, and check that the implied exit multiple is plausible. Run a grid over WACC and terminal growth instead of one point. In Plan Mode you can review forecast logic before execution and keep code plus assumptions on a saved Analysis Page.
4. Horizontal and Vertical (Common-Size) Analysis
Horizontal analysis is the percentage change in each line from a base year. Vertical, or common-size, analysis states each line as a share of a base total, revenue on the income statement or total assets on the balance sheet. Together they show growth, cost mix, and asset mix, and they put firms of different size on one scale.
The CFA Institute refresher on cash-flow statements applies the same idea to cash: each line over revenue, or inflows and outflows as shares of totals. Cash flow to revenue is operating cash flow divided by net revenue. Cash return on assets uses average total assets. Cash return on equity uses average shareholders' equity. Coverage ratios then compare cash with interest, debt repayment, or reinvestment. Strong cash relative to sales can still sit next to heavy capital spending.
Selling, general, and administrative expense rising from 18% to 24% of revenue over five years is a cost-structure story, not a growth story. Two airlines with similar margins can look nothing alike once leased assets dominate one balance sheet.
Use constant-currency figures for cross-border series. Flag percent changes when the base is near zero. Then link the mix shift back to ratios. Charts of each percentage, with the code that built them, are easier to defend than a pasted table. See financial data visualization when the chart is the audit trail.
5. Cost-Volume-Profit and Break-Even Analysis
Cost-volume-profit analysis asks how price, variable cost, fixed cost, and volume set profit. Break-even volume is fixed cost divided by contribution margin per unit. I use it for pricing, capacity, and go or no-go calls, not for valuing the firm.
A subscription at a price of 40 and a variable cost of 22 has a contribution of 18. Divide fixed cost by 18 and you have break-even units. A logistics operator facing a 10% rise in fixed cost does not keep the old volume target. The break-even point moves, and so does the margin of safety.
Don't assume the cost line
Estimate cost behavior with a regression on history. Report fit and residuals. A linear split of "fixed" and "variable" that ignores step costs will give you a break-even that operations can't hit.
- Show operating sensitivity: A high fixed-cost share means profit swings harder than volume.
- Move price and volume together: They interact. A price cut that "only" needs more units may never clear the new break-even.
- Save the function: PlotStudio AI can fit the cost function from your file and keep charts plus code on one analysis page.
Contribution margin is an accounting construct. If variable cost includes allocated overhead, the break-even units are fiction.
6. Sensitivity and Scenario Analysis
Sensitivity analysis moves one input, such as NPV or earnings per share, while the rest stay fixed. Scenario analysis moves several inputs together so the case is a coherent world: downside, base, and upside. Shell's planning work in the 1970s made scenarios a management practice. They are not forecasts.
In 2025, 65% of CFOs reported adjusting financial forecasts and budgets in response to market volatility, according to PwC's Pulse Survey of 678 executives and board members. A grid is how you show what you changed.
A DCF that only prints one value hides the two assumptions that usually dominate, WACC and terminal growth. A utility forecast should tie revenue cases to regulation and demand, not to three arbitrary growth rates. If you also watch how crowds price discrete outcomes, prediction market apps on CoinStats are a different probability surface. They don't replace a documented driver grid.
What to show a reviewer
Rank inputs by effect on the output before you chart them. A tornado chart does that in one view. If you attach probability weights, document them, and don't call the weighted number a forecast.
An unstable cell in the grid is a modeling problem, not a more precise answer.
For a worked scenario analysis layout, run each case, inspect where the output jumps, and tighten the grid only there. PlotStudio AI can do that multi-step pass and keep the cases inspectable.
7. Monte Carlo Simulation
A mining NPV with uncertain commodity prices, operating costs, and a contested discount rate is where Monte Carlo earns its keep. The method draws repeated samples from distributions you specify, computes the output each time, and summarizes the resulting distribution. Stanislaw Ulam and John von Neumann developed it at Los Alamos in the 1940s. Markets adopted it much later, and finance now uses it for project value, portfolio risk such as Value at Risk, pension funding, and options.
A one-year Value at Risk on a multi-asset book is a second natural fit when you use historical bootstrap resampling, which avoids inventing a normal curve the returns don't have.
A 2025 industry survey reported that 90% of respondents were using alternative data, up from 67% the prior year, in the Lowenstein Sandler alt-data report. Extra series don't help a simulation until provenance, timestamps, and leakage controls are written down. Justify each distribution with data or documented judgment, and show how results move when you change that choice.
The short walk-through below is the sampling idea, not a substitute for your own seed log and convergence check.
Fix and report the random seed. Increase iterations until the tails settle. Report the distribution, not only the mean. PlotStudio AI runs the generated Python locally and exports a reproducible Jupyter notebook, so the simulation method stays visible rather than hidden in a chat transcript. Concrete sensitivity analysis examples still belong beside the simulation, because one-way shocks show which inputs deserve a distribution.
8. Multiple Regression and Factor Models
Multiple regression estimates how a financial outcome relates to several drivers at once. Stock returns, default rates, and revenue are the usual dependents. The Fama-French three-factor model (1993) is the applied case most equity work still starts from: market, size, and value. For volatility that clusters, Robert Engle's ARCH (1982) and Tim Bollerslev's GARCH (1986) are the time-series extension. PlotStudio AI can apply GARCH on financial series when the question is volatility, not a static beta.
A fund's factor loadings answer a narrow question. Does alpha survive after market, size, and value? A bank default model against unemployment and rate changes needs Newey-West or clustered errors if residuals are dependent. Pretty coefficients with ordinary standard errors are how that model gets oversold.
Checks before you trust a coefficient
Test for unit roots before you regress levels. Difference, or use a cointegrated specification, when the series aren't stationary. Report the diagnostics. Hold out a later window. Multiple specifications raise data-mining risk, so pre-commit the factors.
Our saved analyses keep assumption checks next to the model choice, which is the review step I want before anyone quotes an alpha. If you need the specification written out first, start from what multiple regression is and only then fit the factors.
9. Event Study Analysis
An event study asks whether a discrete event moved prices beyond what the market would have done anyway. Earnings, mergers, regulation, recalls, and stress-test announcements are the usual events. You fit a benchmark on a pre-event window, compute abnormal returns in the event window, and test whether cumulative abnormal returns differ from zero.
UCL's review of the efficient-market literature places the method in its history. Maurice Kendall (1953) found largely random price changes in 22 stocks. Eugene Fama formalized market efficiency in 1965 and reviewed the evidence in the Journal of Finance in 1970. Ball and Brown (1968) published the first event study. Fama, Fisher, Jensen, and Roll (1969) studied stock splits. Later summaries report that prices typically adjust within about one day, read as evidence for semi-strong efficiency. That speed is not a reason to skip the design. Window, benchmark, sample, and overlapping news still decide the result.
Banks around a stress-test announcement, or a sample of firms disclosing a cybersecurity breach, are workable studies if events don't pile onto the same calendar day. When they do, use a calendar-time portfolio or a cross-sectional correction. Report more than one benchmark. If you scan many windows, adjust for multiple testing. PlotStudio AI can fit the estimation model, compute abnormal returns, and save the event-window plots with the code.
10. Altman Z-Score
The Altman Z-Score folds five ratios into one discriminant score aimed at bankruptcy risk within about two years. Edward Altman's 1968 model for public manufacturers is Z = 1.2X1 + 1.4X2 + 3.3X3 + 0.6X4 + 1.0X5. X1 is working capital to total assets, X2 retained earnings to total assets, X3 EBIT to total assets, X4 market value of equity to total liabilities, and X5 sales to total assets. Scores below 1.81 are the conventional distress zone. Scores above 2.99 are the safety zone. Private-firm and emerging-market variants change both weights and cutoffs.
Academic work on financial statement analysis places the score alongside profitability, liquidity, debt, and activity ratios as one screen among several. That placement is the practical reading. A falling liquidity ratio or rising debt levels can signal pressure while earnings stay positive, and the Z-Score compresses those signals into a single input for triage.
Run it as a workflow. Screen an industrial portfolio and send names below 1.81 to credit review. After a recession, compare score distributions against later defaulters to see whether the old cutoffs still separate. Check the variant before applying 1968 weights to a bank or a private firm. On a holdout sample, report type I and type II errors, then add a market check such as credit spreads. PlotStudio AI runs the Python for each step and saves the code, charts, statistical tests, and interpretation on one Analysis Page, so the score's inputs and results stay inspectable. For wider AI-assisted research data work, see PlotStudio AI.
10 Financial Analysis Techniques at a Glance
| Technique | Primary Question It Answers | Data Required | Key Strength | Main Pitfall |
|---|---|---|---|---|
| Ratio Analysis | How does a firm perform in liquidity, profitability, solvency, efficiency, and valuation? | Financial statement figures and comparable peer or historical data | Fast, transparent way to identify areas for investigation | Accounting choices and one-time items can distort ratios |
| DuPont Analysis | Which factors explain a firm's return on equity? | Net income, sales, total assets, and equity | Reveals whether ROE is driven by margins, efficiency, or leverage | Components are interdependent, so attribution is approximate |
| Discounted Cash Flow (DCF) Valuation | What is the present value of expected future cash flows? | Forecast free cash flows, discount rate, and terminal value assumptions | Links valuation to fundamental cash generation | Small changes in assumptions can cause large valuation swings |
| Horizontal and Vertical (Common-Size) Analysis | How have financial statement items changed over time or as a share of totals? | Financial statement line items across periods and base totals | Makes trends and financial structures easy to compare | Small bases, inflation, or currency changes can mislead |
| Cost-Volume-Profit and Break-Even Analysis | What sales volume is needed to break even or reach a profit target? | Sales price, fixed costs, variable costs, and sales volume | Provides clear, decision-oriented estimates | Assumes linear cost and revenue relationships within a relevant range |
| Sensitivity and Scenario Analysis | How do changes in assumptions affect an outcome? | A model output and input assumptions, varied individually or in scenarios | Makes key drivers and uncertainty explicit | One-at-a-time analysis can ignore correlations between inputs |
| Monte Carlo Simulation | What range and probability distribution of outcomes could result from uncertain inputs? | A model, probability distributions for uncertain inputs, and relevant correlations | Captures interactions and nonlinear effects | Results depend heavily on the quality of chosen distributions |
| Multiple Regression and Factor Models | How are financial outcomes related to several explanatory variables or factors? | A dependent variable and multiple explanatory variables, often over time | Produces interpretable estimates of relationships and factor exposures | Violated assumptions can lead to unreliable conclusions |
| Event Study Analysis | How does an event affect asset prices beyond expected returns? | Asset returns, event dates, estimation and event windows, and a benchmark model | Isolates market reactions to discrete events | Confounding events in the window can bias results |
| Altman Z-Score | How much distress or bankruptcy risk does a firm's financial profile indicate? | Five accounting ratios, with the appropriate model variant for the firm | Simple screening tool for large groups of firms | Historical cutoffs may not transfer across periods or industries |
Choosing, Automating, and Auditing Your Technique Stack
No single method answers every financial question. Ratio and common-size analysis screen. DuPont and cost-volume-profit diagnose. DCF and Monte Carlo value. Regression and event studies test. The Z-Score flags distress. Match the method to the decision, fix the definitions and data prep in writing, and test assumptions before you trust the output.
A model you cannot rerun is a model you cannot defend. Recent work on financial AI, summarized in a Springer Artificial Intelligence Review article, still finds gaps in stable explanations, concept drift, monitoring, and reproducibility. I would rather stake a decision on a simpler technique with clean lineage than a complex one with a black box in the middle. A defensible protocol records source documents, accounting definitions, transformations, model specification, assumptions, software environment, sensitivity checks, and decision thresholds. Restatements, acquisitions, FX translation, segment changes, and non-GAAP adjustments are exactly where ratios drift between periods.
Agentic analytics is data analysis performed by AI agents that can plan an investigation, execute steps with real tools and code, inspect intermediate results, revise the approach, validate outputs, and synthesize a complete analysis. An answer is an output. An analysis is an investigation. In Kyriba's survey of 1,000 CFOs and senior financial decision-makers, 39% of CFOs identified investment analysis as a planned use of AI in 2025. A chat reply evaporates. A rerunnable file holds up in review.
PlotStudio AI is built for analysts and researchers who need to investigate data rather than chat around it. It profiles each dataset on upload, lets you review and edit the plan in Plan Mode, and runs Python locally on your machine. Every result is saved as an Analysis Page that bundles the written interpretation with charts, statistical tests, and the generated code, which you can inspect line by line or rerun as an exported Jupyter notebook or PDF report. Automation here does not remove judgment. It shortens the mechanical path so you spend your time challenging the driver that moved the result.
Download PlotStudio AI and run one of these techniques on a filing extract you already trust: plotstudio.ai.