16 study modules and 22 formulas, from fundamentals through advanced research and execution engineering.
Original study guidance connected to relevant research questions. Reading material is not model training, demonstrated mastery or a profitable trading strategy. No live orders are enabled.
Research algorithm and evidence gates
Evaluate a bounded, cash-only stock strategy after costs and uncertainty; allow an unresolved or negative result.
- mandate: Specify capital, reserve, allowed assets, exposure, loss pause, review owner and benchmark.
Required: Versioned owner-reviewed mandate; no automatic approval from deposit.
- intake: Collect permitted prices, filings and macro observations with point-in-time lineage.
Required: Manifest, timestamps, license, checksums and missing-data report.
- hypothesis: Freeze signal, timing, universe, costs, exits, comparisons and rejection rules.
Required: Registered strategy specification and complete candidate-search log.
- replay: Run chronological out-of-sample replay with realistic costs, gaps and fill constraints.
Required: Reproducible artifacts, matched cash/hold baselines and uncertainty; no invented fills.
- challenge: Check leakage, revisions, survivorship, multiple searches and regime sensitivity.
Required: Independent review; reject or revise when evidence fails the predefined criteria.
- paper: Run prospectively with actual intended capital limits; test broker failures.
Required: Dated paper fills, reconciliation, drawdown and successful fault-injection reports.
- release: Review evidence, implementation, funding, account permissions and monitoring.
Required: Separate authenticated owner activation; this educational kit cannot activate trades.
- monitor: Track positions, costs, flow-adjusted performance, stale data and policy changes.
Required: Audit trail, tested alerts, pause/recovery process and scheduled human review.
acceptance: Predeclare numeric performance, cost, drawdown, sample-size and uncertainty thresholds before the final test. No universal profitable threshold or number of trades is asserted here. Require all artifacts and independent review; cash is a valid outcome.
rejection: Reject a candidate if it fails the preregistered benchmark/cost/risk rules. Revise the method when data or tests are insufficient; do not relabel missing evidence as success.
safety: No margin, shorting, derivatives, live orders, transfer, credential handling or automatic activation in this kit. Halting new entries does not cap losses in existing holdings.
replication: Preserve source manifests, versions, split boundaries, seeds, parameters, costs, all trials and output hashes. A second reviewer must reproduce the result.
next state: retain as an educational and research workflow; trading readiness unresolved
boundaries: Logical role mapping, not a claim that new live data connectors or autonomous workers were provisioned. Shared research context receives selected notes, not all textbooks or model training.
Connections between matrix roles
Finance and economics: dated filings, prices and macro releases → testable hypotheses and valuations
Statistics and machine learning: point-in-time features → held-out metrics, uncertainty and calibration
Simulation: frozen rules, costs and data manifests → labeled simulated fills and stress results
Evidence review: claims and reproducible artifacts → supported, contradicted or unresolved decisions
Operations: reviewed specification → readiness checks and audit requirements; no authorization from AI text
Curriculum
Foundation · mastery not assessedStocks, funds and market risk
Distinguish ownership, dividends, price appreciation, diversification and total loss. A share price alone does not measure value.
Assessment: Explain why a cheaper share is not necessarily a cheaper business.
Prerequisites: none
Reading: SEC Investor.gov: Stocks
Foundation · mastery not assessedFinancial statements
Connect assets = liabilities + equity, cash flows, earnings, debt and dilution. Cash and accounting profit differ.
Assessment: Reconcile a company balance sheet and operating cash flow using dated filings.
Prerequisites: ownership
Reading: SEC EDGAR APIs · OpenStax Principles of Finance
Undergraduate · mastery not assessedReturns and time value
Separate price return, dividends, fees, taxes, deposits and withdrawals. Compare identical horizons and currencies.
Assessment: Calculate net total return without counting deposits as profit.
Prerequisites: ownership
Reading: OpenStax Principles of Finance
Undergraduate · mastery not assessedValuation and uncertainty
Discount scenario cash flows; vary growth and discount rates. P/E is undefined or misleading with nonpositive earnings; valuation is not a timing signal.
Assessment: Produce pessimistic, central and optimistic cash-flow assumptions with sensitivity analysis.
Prerequisites: accounting, returns
Reading: OpenStax Principles of Finance · SEC EDGAR APIs
Undergraduate · mastery not assessedProbability and estimation
Study sampling error, dependence, outliers, covariance and base rates. An estimated probability is not a known probability.
Assessment: Compare mean returns and uncertainty using a dependence-aware resampling method.
Prerequisites: returns
Reading: OpenStax Principles of Finance
Undergraduate · mastery not assessedPortfolio construction
Account for correlated exposures, liquidity, concentration and estimation error. Diversification does not remove market risk.
Assessment: Compare a cash baseline and diversified benchmark at matched exposure.
Prerequisites: statistics
Reading: OpenStax Principles of Finance
Undergraduate · mastery not assessedQuotes, orders and liquidity
Bid, ask, spread, timestamps, venue coverage, depth and partial fills affect realizable results. A displayed quote is not a guaranteed fill.
Assessment: Reconcile one order from submission through partial fills, cancellation and fees.
Prerequisites: returns
Reading: Alpaca order documentation
Undergraduate · mastery not assessedPoint-in-time data
Record source, license, observation time, publication time, retrieval time, revisions, adjustments and checksum. Avoid using later revisions in earlier decisions.
Assessment: Build a data manifest and demonstrate that every feature existed before its decision.
Prerequisites: accounting, statistics
Reading: SEC EDGAR APIs · Federal Reserve FRED API
Graduate · mastery not assessedTestable trading hypotheses
Specify universe, signal, decision time, holding period and exit before testing. Include no-trade when expected benefit cannot justify estimated costs and uncertainty.
Assessment: Freeze one candidate configuration and record every alternative tried.
Prerequisites: provenance, microstructure
Reading: scikit-learn TimeSeriesSplit
Graduate · mastery not assessedChronological evaluation
Separate development, validation and untouched final periods. Purge overlapping labels, add a justified gap, include delisted securities and model fills conservatively.
Assessment: Reproduce a walk-forward replay including splits, dividends and stressed costs.
Prerequisites: signals, portfolio
Reading: scikit-learn TimeSeriesSplit
Graduate · mastery not assessedMachine learning and calibration
Fit scalers and feature selection only inside training folds. Evaluate calibration, drift, stability and ablations; compare with simple rules. LLM explanations are not price labels.
Assessment: Show an untouched test, calibration plot and performance without each feature family.
Prerequisites: backtesting
Reading: scikit-learn TimeSeriesSplit
Graduate · mastery not assessedEconomic regimes and global comparison
Align release calendars, currencies and historical vintages. Treat regime labels as uncertain; worldwide survey sentiment is not a causal stock forecast.
Assessment: Repeat a macro-feature test using only information released at each historical date.
Prerequisites: provenance, portfolio
Reading: Federal Reserve FRED API
Graduate · mastery not assessedFactors, derivatives and stochastic models
Study factor regressions, duration, convexity and option models with explicit assumptions. Mathematical pricing models do not establish a tradable edge or authorize leverage.
Assessment: Check limiting cases and sensitivity of an advanced model; keep derivatives outside the initial cash pilot.
Prerequisites: valuation, statistics
Reading: OpenStax Principles of Finance
Research · mastery not assessedIndependent challenge and replication
Freeze data hashes, code, parameters, benchmark, costs and acceptance rules. Report failed candidates and all searches. Reproduction is distinct from independent replication.
Assessment: Have another reviewer rerun the package and challenge leakage and cost assumptions.
Prerequisites: ml, macro
Reading: scikit-learn TimeSeriesSplit
Engineering · mastery not assessedExecution reliability and recovery
Use durable client identifiers, account-wide reservations and reconciliation. An order timeout means unknown status, not permission to resubmit. Test disconnects and cancel races.
Assessment: Inject duplicate events, partial fills and lost acknowledgements; prove no duplicate exposure.
Prerequisites: microstructure, backtesting
Reading: Alpaca order documentation
Engineering · mastery not assessedMonitoring and controlled release
Version strategy and limits; record settled cash, external flows, holdings, costs and drawdown. Pause new entries on stale data or reconciliation failures. Review activation independently.
Assessment: Demonstrate halt, alert delivery, recovery and rollback without sending a live order.
Prerequisites: operations, replication
Reading: Alpaca order documentation
Formulas and limits
Original mathematical reference notes; these are not trading signals or measured outcomes.
total-return
R=(P1-P0+D)/P0P0,P1: comparable share prices; D: dividends per share; R: dimensionless.
P0>0; exclude external cash flows; do not add dividends again to a total-return adjusted series.
log-return
r=ln(1+R)R: total return; r: log return, dimensionless.
R>-1; sums across time, not across portfolio weights.
compound
Vn=V0 × product(1+Rt)V: currency; Rt: each period return.
No external flows; historical returns are not promised future rates.
present-value
PV=sum(CFt/(1+k)^t)CFt: currency at period t; k: discount rate per period.
k>-1; cash flows and discount rate must share currency, inflation basis and horizon.
npv
NPV=-I0+sum(CFt/(1+k)^t)I0: initial currency outlay; remaining symbols as present value.
A forecast under assumptions, not evidence of execution or market price.
cagr
CAGR=(Vn/V0)^(1/Y)-1Vn,V0: flow-adjusted values; Y: years.
Positive endpoints and Y>0; hides interim losses.
mean
mean(R)=sum(Rt)/nn: number of comparable periods.
Historical arithmetic average; serial dependence affects inference.
variance
s²=sum((Rt-mean(R))²)/(n-1)Returns: dimensionless; n>1.
Sample variance; not a complete tail-risk measure.
portfolio-return
Rp=sum(wi Ri)wi: beginning-period portfolio fraction; Ri: same-period returns.
Include cash; weights sum to one for unlevered fully accounted portfolio.
portfolio-risk
variance(Rp)=wᵀ Σ wΣ: covariance matrix for same-horizon returns.
Estimated correlations can change; covariance matrix must be valid.
sharpe
S=mean(Rp-Rf)/sd(Rp-Rf)Rf: same-period risk-free return; S: dimensionless.
Nonzero standard deviation; sqrt(periods/year) annualization requires suitable dependence assumptions.
beta
beta=cov(Ri,Rm)/var(Rm)Ri: asset returns; Rm: market returns.
Nonzero market variance; descriptive sensitivity, not causality.
capm
E[Ri]=Rf+beta_i(E[Rm]-Rf)Expected returns: common horizon, dimensionless.
Equilibrium model with restrictive assumptions; not a guaranteed return forecast.
drawdown
DDt=1-Et/max(E0,...,Et)E: positive flow-adjusted equity or unitized index.
External deposits must not hide losses; daily sampling misses intraday drawdown.
spread
spread_bps=10000(ask-bid)/((ask+bid)/2)ask,bid: currency/share; bps: basis points.
Positive contemporaneous noncrossed quotes; venue coverage must be stated.
net-pnl
net_PnL=q(Psell-Pbuy)+dividends-fees-other_costsq: shares; prices: currency/share; result: currency.
Use actual fills; do not subtract spread twice if already reflected in fills. Taxes reported separately.
flow-adjustment
period_PnL=ending_equity-starting_equity-deposits+withdrawalsAll terms in the same currency and interval.
For percentage returns with intraperiod flows use unitization or a justified cash-flow timing method.
expected-payoff
EV=pG-(1-p)L-Cp: estimated win probability; G,L,C: currency gain, loss, costs.
All estimates uncertain; positive sample EV does not establish future profit.
sma
SMA_n(t)=sum(P[t-i], i=0..n-1)/nP: comparable price; n: bars; result: currency/share.
A close-based signal cannot assume execution at that same already-known close.
budget
order_cap=max(0,min(order_limit,exposure_room,settled_cash-reserve-open_buy_reservations))All terms: currency; account-wide reservations.
Illustrative cap; enforce asset precision, fees, price movement and broker rules before any order.
kelly
f*=(bp-(1-p))/bb: net win/loss payoff ratio; p: assumed win probability; f*: fraction.
Ideal repeated independent binary bets with known probabilities; unsuitable as automatic sizing for uncertain stock forecasts. Study only.
tracking-error
TE=sd(Rp-Rbenchmark)Same-horizon matched portfolio and benchmark returns.
State sampling interval; assess costs and exposure differences before comparing.