Financial research · models and dataRESEARCH / DEMO
PERSONAL SOFTWARE & DATA SCIENCE PROJECT
Financial research, built end to end.
From SEC filings to tested statistical models and a private portfolio research platform. QuantForge brings data engineering, quantitative analysis, financial modeling, and secure application development into one project.
Inside the project
Four connected disciplines
01 · DATA ENGINEERING
Source-backed statements
SEC Company Facts ingestion, fiscal-period selection, filing lineage, missing-value controls, and reproducible exports.
02 · QUANTITATIVE ANALYTICS
Risk and optimization
Covariance shrinkage, allocation constraints, downside simulation, and risk-return comparisons with clear modeling assumptions.
03 · STATISTICAL RESEARCH
Validation beyond backtests
Rolling out-of-sample evaluation, fees, benchmark comparisons, robustness checks, and integrity fingerprints in the private research system.
The public models here are intentionally accessible; the authenticated research engine is substantially deeper. Its methods, algorithms, validation design, and architecture are documented publicly without exposing personal research sessions or licensed market data.
RESEARCH ENGINEWalk-forward testing
Past-only training, non-overlapping evaluation, costs, passive and rebalanced benchmarks.
Statistical validation
PORTFOLIO SCIENCEConstrained optimization
Covariance estimation, weight caps, risk diagnostics, alternative allocation rules.
Optimization + risk
FUNDAMENTAL RESEARCHSEC scenario forecasting
Reported annual filings, FCF proxies, discounting, terminal-value and sensitivity analysis.
Financial modeling
PRODUCT ENGINEERINGAuthenticated analytics
Read-only market feeds, owner-only storage, MFA, content security, and CI.
Annual financial statements, trends, and comparative metrics sourced from SEC EDGAR.
Explore a different sector:
Company analysis
SEC annual filings
Annual revenue trend
Reported · fiscal year
Annual reported figures
Company comparison
Latest available fiscal year
Rankings compare the latest available annual filing for each issuer, not a synchronized period or a sector-neutral investment signal. Missing values are unranked.
Inspect source tags and filing provenance
Source: SEC EDGAR Company Facts.
SEC company financials are real, historical disclosures. The margin scenario is hypothetical. Public portfolio, Quant, DCF and Company peers use illustrative or simulated inputs; they are not actual market results, valuations, or investment recommendations.
Portfolio simulator
A sample portfolio with simulated holdings and prices. Make hypothetical transactions in this browser session; no orders or real-time market quotes.
Total value—
Cash—
Holdings—
Positions
Virtual transaction
Assumed 0.05% transaction fee; no real orders
ETF holdings overlap
Illustrative ETF weights, not actual fund holdings
Rebalancing plan
Target weights sum to 100% including cash. Suggestions only; no transactions.
Quant
Constrained allocation and downside scenarios using illustrative returns.
Analysis settings
276 simulated sessions · 1,800 Monte Carlo paths
Valuation
Interactive discounted cash flow methodology using illustrative company assumptions. These values are not actual company valuations; SEC Financials contains the real reported figures.
Company DCF
Company peers
Compare company characteristics using standardized model inputs. The SEC Financials tab provides reported company figures.
Five features: revenue growth, operating margin, net margin, free-cash-flow margin and debt/assets.
RESEARCH SYSTEMS · PUBLIC TECHNICAL EXHIBIT
Under the hood of the research engine.
QuantForge combines SEC data engineering, constrained portfolio mathematics, leakage-aware historical evaluation, financial modeling, and authenticated application design. This is a transparent technical explanation of the owner-only capabilities—not access to private data.
Real SEC disclosures in public explorerIllustrative diagrams belowPrivate datasets stay private
One platform, four computational stages
Architecture at a glance
01 / DATA ENGINEERINGIngest & normalizeSEC annual facts; Alpaca historical equity, ETF and supported crypto observations in the private workspace
02 / VALIDATIONTrust the inputsFiscal periods, market calendars, aligned prices, missing values, source attribution and SHA-256 dataset fingerprints
03 / MODELSQuantify trade-offsPortfolio risk, constrained allocation, scenario DCF proxies, historical-momentum research and stress tests
04 / EVALUATIONChallenge the resultPast-only fitting, out-of-sample windows, fees, passive benchmarks, leave-one-window-out robustness and research reports
The public deployment hosts static demonstrations and a reviewed SEC snapshot. Market observations and personal virtual accounts are only accessed in authenticated private workflows.
Technical capability walkthrough
Algorithms · diagnostics · engineering judgment
QUANTITATIVE FINANCE / NUMERICAL OPTIMIZATION
01. Portfolio optimization is a constrained mathematical problem
Allocation weights are fitted from observed returns and an estimated covariance matrix, subject to feasibility and concentration limits.
Implemented privately
min wᵀΣw subject to Σwᵢ = 1; 0 ≤ wᵢ ≤ cap
Why it matters: the objective is predicted portfolio variance, not a guarantee of realized return. The private optimizer supports a covariance-based minimum-risk allocation and alternative inverse-volatility / return-tilt allocations.
Align price histories to comparable observations before estimating returns.
Apply covariance stabilization/shrinkage and reject impossible weight caps.
Compare annualized volatility, drawdown, tail loss and alternative allocations.
Implementation and verification detail
Estimated portfolio volatility is √(wᵀΣw) × √annualization; optimization remains sensitive to return sampling, correlated assets and constraints. Tests check symmetry, finite prices, positive covariance diagonals, feasibility and weight normalization.
Schematic risk vs. return trade-off
Illustrative positions, not an empirical efficient frontier or a private portfolio result.
STATISTICAL VALIDATION / TIME SERIES
02. Walk-forward research tests future windows, not fitted history
A model trains on past observations, then is evaluated on the following unseen sessions. The allocation is refitted at later boundaries.
Implemented privately
Train[t₀:t₁] → Fit w₁ → Test[t₁:t₂] → Roll forward
The private study supports non-overlapping evaluation windows, modeled turnover and basis-point transaction fees, two rebalanced strategies, and a passive equal-weight buy-and-hold comparison.
Run each allocation rule against matched evaluation dates.
Reconcile per-window fees to total reported costs.
Perform leave-one-window-out diagnostics; a changed winner signals sample dependence.
Preserve observations and model settings in versioned research manifests.
Where bias can still enter
Past-only fitting reduces mechanical look-ahead within the test, but cannot eliminate survivorship bias, post-hoc asset selection or tuning parameters after seeing outcomes. Omitting a window without refitting is a sensitivity check, not independent validation.
Rolling evaluation boundaries
Window 1
Window 2
Model portfolioBenchmark
Diagram only; wealth paths are drawn for explanation, not copied from private analysis.
FINANCIAL MODELING / SEC DISCLOSURES
03. Forecast sensitivity starts with reported facts
Observed company financials are kept distinct from hypothetical revenue-growth, cash-flow-margin and discounting assumptions.
Real source + modeled scenarios
FCFₜ = Revenue₀(1+g)ᵗ × margin
PV = Σ FCFₜ/(1+r)ᵗ + TV/(1+r)ᴺ
Real SEC annual statements establish the observed baseline. The model projects three- to seven-year hypothetical scenarios and produces a sensitivity grid for revenue growth and the discount rate.
Report free cash-flow proxy as operating cash flow less cash PP&E purchases.
Distinguish discounted cash-flow proxy from an equity value or share-price target.
Display how much present value comes from the terminal component.
Reject impossible assumptions and missing SEC cash flows instead of fabricating data.
Terminal-value risk
Terminal value extrapolates cash flows beyond the explicit projection horizon. When the terminal share dominates, small changes in discount and perpetual growth rates can materially change the result. Debt, excess cash and capital-structure adjustments are not included in this proxy.
How scenarios respond to assumptions
Change
Effect on modeled proxy
Higher assumed revenue growth
Usually higher
Higher assumed FCF margin
Higher
Higher discount rate
Lower
Higher terminal growth (r > g)
Usually higher
Directional sensitivities are mathematical examples, not measured company forecast accuracy. Real reported values are available in SEC Financials.
DATA SCIENCE / EXPLAINABLE RESEARCH
04. Investment ideas rank evidence, not certainty
A private shortlist compares price momentum, realized volatility and drawdown, while displaying available SEC fundamentals separately.
Owner-only research
Score = Σ (lens weightₖ × normalized metricₖ)
Growth, balanced and risk-aware lenses change how much weight the model places on trailing return, recent momentum, annualized volatility and historical drawdown. The score is an explainable heuristic, not a probability of gain.
Use selectable investment names rather than requiring ticker memorization.
Flag sparse histories and report the period and IEX/SIP source.
Show SEC revenue growth and FCF margins only when disclosed facts support them.
Keep ranking and trade execution separate; no brokerage orders are placed.
Why a top-ranked investment may still be unattractive
Momentum and low observed volatility do not measure whether a stock is expensive relative to its future cash flows. Future returns also depend on valuation, concentration, news and unforeseeable events. The research lens is for comparing candidates, not picking guaranteed winners.
Explainability layers
EvidencePrice history, volatility, drawdown and optional SEC context
ScoreWeighted, bounded and traceable to its inputs
DecisionManual research and hypothetical portfolio workflow, not an order
The private browser uses real historical observations when authorized; no private rankings or selected holdings are displayed here.
Research integrity and engineering controls
Reproducibility + security
Source lineage
Issuer and fiscal-year filtering, filings and cutoffs, calendar alignment, adjusted-provider-price attribution and unavailable-value handling.
Reproducible manifests
Versioned exports include investment symbols, settings, evaluation dates, fee rules and SHA-256 input fingerprints; raw licensed prices are not exported.
Testable calculations
Python and JavaScript tests include invalid prices, impossible allocation caps, missing filings, look-ahead prevention, fee reconciliation and prior-window comparisons.
Release gates
GitHub Actions checks Python 3.11 and 3.12, the private regression suites, privacy review, script syntax and compiled Content Security Policy hashes.
PUBLICCloudflare PagesReal SEC snapshot and labeled browser-only illustrative models
EDGEScoped API proxyRestricted configured upstream; private backend credentials not returned
PRIVATE APIRender / FastAPIAuthorized requests for market observations and personal state
IDENTITY & DATASupabase MFAAuthenticator step-up and owner-scoped virtual account records
The diagrams document intended and tested boundaries, not a claim that any production system is immune to attack. SHA-256 fingerprints detect changed in-memory inputs but do not authenticate the provider or replace archived source data.
TRY THE PUBLIC SIDE
Methods and interactive examples, without sign-in
Compare real annual company filings in SEC Financials, run illustrative optimization and valuation models, or inspect the data and test methodology. Advanced live market-data research remains authenticated.
METHODS
Methods & engineering
Data sources, analytical methods, assumptions, and application architecture.
Advanced private research, publicly documented
Implementation walkthrough
The authenticated workspace implements walk-forward optimization, volatility and tail-risk diagnostics, investment shortlisting, discounted cash-flow scenario grids, and research integrity checks. The Research systems page shows equations, validation boundaries, explanatory diagrams and exact analytical assumptions—without a private sign-in.
The public browser does not request raw filings or private market quotes. A locally generated and reviewed snapshot is embedded when the site is built. Figures do not auto-refresh.
Missing values remain unavailable rather than zero-filled. Company fiscal year-ends are not aligned; broad sector groupings are editorial and are not normalized industry classifications.
Model demonstrations and boundaries
Interactive · illustrative
Portfolio simulator
Virtual positions, example prices, rebalancing weights, and estimated fees. Everything stays in the browser session and does not place orders.
Quantitative modeling
Simulated returns, covariance, constrained allocation, and Monte Carlo downside. Results are not calibrated forecasts of market risk.
DCF sensitivity
Illustrative cash-flow assumptions with discounting and terminal-value sensitivity. These are not live company valuations.
Company similarity
Standardized synthetic fundamentals demonstrate nearest-neighbor comparisons. Use SEC Financials for actual company comparisons.
Private API credentials stay server-side. No Alpaca observations, personal positions, or authenticator secrets are bundled into the public demonstration. Research scripts have deterministic tests, and GitHub CI validates both Python and JavaScript with a restrictive Content Security Policy.
Engineering practices
Architecture and tests
Data engineering
Named SEC issuer identifiers, annual-period filtering, filing cutoffs, input validation, recorded sources and accession details.
Python backend, static hosted frontend, automated Python and JavaScript checks, continuous integration, and a restrictive script Content Security Policy.
Privacy boundary
Public research data is separate from the owner-only workspace. Private holdings and Alpaca access are protected by authenticated endpoints and authenticator MFA.
The private workspace provides real market observations and virtual trading for the owner only. Public visitors cannot use that API as an open data source.
How the research is tested
Quality gates and failure cases
Layer
Example invariants
Why it matters
Historical data
Finite positive prices; aligned time series; observed-versus-derived SEC metrics
Rejects inconsistent or fabricated model inputs
Optimization
Weights sum to one; asset caps feasible; covariance symmetric; finite estimates
Prevents invalid portfolios from appearing plausible
Walk-forward
Past-only fitting; disjoint test windows; transaction-fee reconciliation
Reduces leakage and makes comparisons auditable
Forecast
Growth and discount assumptions valid; unavailable FCF remains unavailable
Tests confidentiality and deployed script integrity
CI runs Python 3.11 and 3.12 along with JavaScript checks. Passing tests provide scoped evidence, not investment performance claims or complete security assurance.
Historical financial disclosures may be updated or restated. Model scenarios are not investment advice.
My workspace
Private · Sign in required
Sign in
Only the authorized owner account can sign in.
Authenticator verification
Enter a six-digit code from your authenticator app.
Scan the code with an authenticator app. Keep the authenticator and its recovery options secure.
Personal research · authenticated
Research investments
Shared by Market research, Optimization, Backtesting and Walk-forward
Choose 2–5 investments by name, choose a historical period, and load the prices once. All four research views use that same loaded dataset.
Choose investments to compare.
Selected investments:None selected
Loaded for analysis:No historical data loaded
Select investments, then load history to activate the analysis tools.
Virtual total—
Available cash—
Invested value—
Holdings
Portfolio concentration
Add virtual holdings to see allocation.
Market observation & virtual trade
Stocks, ETFs, and supported crypto pairs. Market observations are private and read-only; no brokerage orders are placed.
No investment selected
No quote loaded.
PERSONAL RESEARCH · OBSERVED DATA
Investment ideas
A plain-language research shortlist built from private historical market observations and, where available, SEC company filings. The ranking is a transparent screening heuristic, not a return forecast, buy signal, or individualized advice.
Companies to examine
Choose companies by name or use the starting shortlist.
How to interpret this
A higher research score indicates a more favorable balance of past momentum and realized risk under the selected lens. It is not an estimated probability of profit, expected return, or price target. SEC results describe past business finances separately; ETFs generally do not have comparable company income statements. Reassess against valuation, diversification, fees, tax impact, investment horizon, and the risk of loss before any real-money decision. IEX-only bars are not consolidated market-wide prices.
Historical market research
Historical daily stocks, ETFs, and supported USD cryptocurrencies from the private Alpaca connection. Stock prices exclude dividends; crypto trades every day.
The investments and date range currently loaded for this analysis are shown in Research investments above.
Load a historical series to begin.
Portfolio optimization
Use the historical prices selected in Market research to explore constrained, long-only allocations. The research calculation stays separate from your virtual positions. Nothing is traded or saved to your portfolio.
Select 2–5 investments in Market research, load their history, then calculate an allocation.
Risk diagnostics
Empirical covariance uses 20% diagonal shrinkage. Historical correlations, contribution to variance, drawdowns, and tail losses are descriptive, not forecasts.
Risk diagnostics appear after an allocation is calculated.
Historical strategy simulator
Equal-weight buy-and-hold versus monthly rebalancing, using the symbols and daily history selected in Market research. Includes hypothetical transaction fees; no forward-looking optimization.
Walk-forward strategy validation
Fit portfolio weights using previous historical observations, then evaluate the allocation on later dates. Refit at every test-window boundary. Compare against equal-weight rebalancing using the same transaction-fee assumptions. No brokerage orders.
Load historical investments in Market research to begin.
Evaluation diagnostics
Strategy robustness and benchmarks
Compare a passive buy-and-hold benchmark, alternative allocation methods, higher fees, and leaving out each evaluation window. All allocation-rule comparisons hold the evaluation dates fixed. For longer tests, select 2 years in Research investments and reload history first.
Run Walk-forward first, then check robustness.
Model uses price-only historical returns; no dividends, borrow costs, slippage, taxes, or corporate-action simulations. Backtests remain subject to selection bias and do not establish future performance.
Financial valuation scenarios
Historical company financials from SEC filings, with editable forecast assumptions. Alpaca supplies market prices, not audited accounting statements.
No company selected
Multi-year SEC financial projections
Choose a company above and load its SEC annual disclosures first. Enter hypothetical annual revenue growth and free-cash-flow margin; projections never overwrite reported financials. Scenarios differ by ±4 percentage points of growth and ±3 points of FCF margin.
Load SEC statements for a selected company first.
Forecast assumption sensitivity
Selected company comparison
Virtual account balance
Change your simulation budget without any bank transfer or brokerage order. Deposits and withdrawals affect only uninvested cash. Reset clears virtual positions and trades but preserves notes.