We keep formulas separate from source data, user assumptions and AI-generated textual interpretation. A calculation runs only when all required inputs are available.
STMR · 1.0.0 · experimental methodology
Technical market classification
The trend score describes a technical state based on closed daily prices and momentum. In the Research overview, it carries a 70% weight alongside the 30% fundamental background when 30% of the original weight is available from at least two of its dimensions. The technical market dimension is excluded here to avoid double counting. Fundamentals do not compensate for a missing technical picture. This is not a probability of returns, profit or suitability. The weights and thresholds are design assumptions; the method’s predictive capability has not been validated.
Data and eligibility
Accurate quote, exchange and currency; consistently provider-adjusted OHLC; 260 consecutive closed observations. We work from the existing research, without new provider calls. The initial profile applies to equities and the verified broad equity ETF list. Crypto, derivatives, illiquid, leveraged and inverse products are excluded until separately validated.
A · Long-term trend · 30%
Average[sign(close − SMA200), sign(SMA50 − SMA200)]. SMA is the arithmetic average of the specified number of observations.
B · Medium-term momentum · 25%
Average[sign(Ct/Ct−63 − 1), sign(Ct/Ct−126 − 1)]. Observed price returns, not total returns with dividends reinvested.
C · Short-term momentum · 20%
Wilder RSI14: below 30 −0.5; 30 ≤ RSI < 45: −1; 45 ≤ RSI < 56: 0; 56 ≤ RSI ≤ 70: +1; above 70: +0.5. The average of this value and the sign of the MACD(12,26,9) histogram. EMA: SMA seed for the first period, then a 2/(n+1) weighting. RSI: average gain/loss over the first 14 changes, then Wilder smoothing.
D · Trend alignment · 15%
Average[sign(SMA20 − SMA50), sign(SMA20t − SMA20t−5)]. A zero difference scores zero.
E · Volume confirmation · 10%
Average volume over the last 5 days / average volume over the preceding, non-overlapping 20 days. At a ratio of at least 1.2, the sign of the five-day price return; otherwise 0. If volume is incomplete or zero, E is omitted; the weights of A–D are normalized from 90% to 100%. If any group among A–D is missing, there is no score.
Summary and labels
100 × weighted group average. ≥60: Strong bullish; ≥20 and <60: Bullish; >−20 and <20: Neutral; >−60 and ≤−20: Bearish; ≤−60: Strong bearish. Classification uses unrounded values. Missing data is not Neutral.
Risk, separately
Sample standard deviation of 20-day log returns × √252 × 100; maximum drawdown from a peak over 252 observations; Wilder ATR14 / close × 100. They do not modify the trend score. ATR is calculated from the true high/low and previous close.
Calendar, freshness and comparison
Only days already closed at the time of research; a 30-minute data-arrival buffer after the exchange close. Weekends and known exchange holidays are not missing. Currently checked against the US/XETRA calendar from 2025-09-01 to 2026-12-31. No classification is available for another calendar or incomplete continuity. Comparisons use common closed days; original data dates are retained.
Recalculation and retention
On request, using retained research data and the original data date. The cache is identified by the complete input hash, exact listing, currency, adjustment, source revision, closing date and method version. An adjustment or methodology change receives a new cache key. Public disclosures are kept in a separate, hashed, append-only log, distinct from internal trials.
Validation status
The formulas and edge cases can be tested automatically; this is not economic forecasting validation. Historical analysis can be interpreted using chronological train/validation/test separation, point-in-time data, uncertainty intervals and simple trend/momentum baseline models. Subsequent provider revisions to the adjusted data series and survivorship bias in today’s universe are limitations.
Scientific methods
Each card explains what the method is, why it is used, what data it requires, what it calculates, how it can be interpreted and what it is not suitable for.
Asset research
Multifactor profile
Multi-factor profile
What is this?
Comparison across six complementary company and market factors.
Why is it used?
Shows the relative position of value, quality, growth, investment, momentum, and risk separately.
What can it be applied to?
equity
Which data are needed?
comparable companies · at least 5 suitable peers
What does it calculate?
For each factor, the average percentile rank of the valid metrics; for metrics where a lower value is favorable, the rank is reversed.
How should you interpret it?
Relative profile, not expected return and not a buy signal.
What are its limits?
The choice of peer group and accounting definitions can materially change the rank.
Minimum data
At least three valid factors, at least 5 suitable peers per factor, and 35% available weight.
Building a financial strength profile from nine accounting signals.
Why is it used?
Shows which profitability, leverage, and operating signals are met.
What can it be applied to?
equity
Which data are needed?
két összehasonlítható annual report · cash flow and balance sheet data
What does it calculate?
ROA>0, CFO>0, ΔROA>0, CFO>net income, Δleverage<0, Δcurrent ratio>0, no dilution, Δgross margin>0, Δasset turnover>0; the number of signals met ranges from 0–9.
How should you interpret it?
The score represents the number of signals, not a return forecast.
What are its limits?
Not applicable to banks, ETFs, or non-comparable periods.
Minimum data
Comparable original data is required for all nine signals.
Alongside the threshold loss, also shows the average loss beyond the threshold.
What can it be applied to?
equity · bank · reit · etf · crypto · bond · fund · fx · commodity
Which data are needed?
portfolio return series
What does it calculate?
VaRα is the α-quantile of losses; ESα is the average loss in the worst (1−α) fraction; the boundary observation receives a fractional weight when necessary.
How should you interpret it?
Displayed as a positive loss percentage.
What are its limits?
Historical frequency, not a future probability; few tail observations are unstable.
Minimum data
At least 100 returns for 95%, and at least 300 returns for 99%.
rₜ = ln(Pₜ / Pₜ₋₁); Vₜ = V₀ × exp(Σ r*). Returns marked with an asterisk are randomly selected, contiguous blocks of historical log returns. Overlapping blocks can be selected with equal probability; the final block may be truncated at the end of the horizon.
Inputs
At least 60 ordered, unique, positive daily closing prices; source, currency, adjustment basis, and original data date. The interface allows 1 000 / 5 000 / 10 000 paths, 21–365 time steps, starting capital, and an integer seed to be set. The block length here is 5 days; the server model accepts 1–20 days.
Assumptions
The observed return blocks may be reused in the experiment being analyzed. Order within each block is preserved; the model does not preserve longer-range dependence between blocks. The fixed seed, data fingerprint, and parameters produce a reproducible result.
Limitations
Not a calibrated forecast. It does not infer shocks, regime changes, or liquidity missing from the past. There are no fees, taxes, currency conversions, or separate dividend credits. The simulated ratio is not a proven future probability. The percentile uses linear interpolation at position (N−1)×q in the sorted sample; it is not a confidence interval.
Backtest · historical testing of rules
Formula
Total return = Vend / Vstart − 1. Annualized return = (Vend / Vstart)^(1/year) − 1. Sharpe = (average daily return − daily risk-free return) / sample standard deviation of daily returns × √A, where A=252 or 365. Daily risk-free return = (1 + annual rate)^(1/A) − 1.
Inputs
Valid daily open, high, low, close, and volume. RSI period and entry threshold, relative volume, stop, profit target, holding days, position size, starting capital, fee, basis-point slippage, and annual risk-free rate.
Assumptions
Entry signal: RSI ≤ selected threshold, relative volume ≥ selected minimum, and closing price ≥ SMA50 × 0,92, jointly. The interface uses fixed RSI14, 10 000 starting capital, and 100% position size. Signal at the daily close, entry at the next observed open. Entry price × (1+slippage), exit price × (1−slippage); the fee is charged to capital on both sides. For a gapped stop, the less favorable opening price applies; if the stop and target are both hit on the same candle, the stop is executed first. The benchmark is holding the same instrument without fees.
Limitations
There is no intraday event sequence, market impact, or execution guarantee. Past performance does not validate a future strategy; testing many parameters may lead to overfitting. Annualization assumptions and the time dependence of returns limit the interpretation of Sharpe. With zero standard deviation, Sharpe=null; with zero trades, win rate=null.
Research Score · proprietary, fixed research index
Formula
Five dimensions at 20% each: valuation, quality, growth, financial stability, and market condition. Within each dimension, metric weights are equal. Total score = Σ(scoreᵢ × weightᵢ) / Σ(available weightsᵢ). Scale values are linearly interpolated; valuation requires percentiles from at least five comparable issuers. Other monotonic metrics are also converted to percentiles when at least five suitable peers are available; otherwise, the documented fixed scale is used.
Inputs
P/E, P/B, free cash flow yield; ROE; revenue and EPS growth; total debt / equity; RSI14, price/SMA50 deviation, and annualized volatility. Each value includes its source, original date, unit, and period. The detailed score view shows the specific scales and comparables. For verified annual SEC metrics, the as-of date and filing date are retained; we accept an as-of date no more than 455 days old, with source verification within seven days. For price-linked and conventional provider snapshots, the seven-day data limit remains. Annual and TTM values cannot substitute for one another.
Assumptions
A higher score on the documented editorial scale indicates a more favorable data profile. The weights and thresholds are product decisions, not literature-validated optimal investment weights. At least five valid metrics, 50% weighting, and two fundamental dimensions are required; with partial data, coverage is shown separately.
Limitations
No validated predictive performance. The total score is not a price target, return, or buy signal. Missing data is omitted; a genuine comparability conflict blocks the result. A bank, ETF, or other asset cannot inherit the company model; the separate bank/ETF profile version and its data gaps stand alone. The sources and the independence of the underlying upstream data are separate questions.
Scenarios · asset-type-specific formulas
Formula
Company: conditional price = EPS₀ × (1+g)^(h/12) × terminal P/E. Bank: conditional price = BVPS₀ × (1+g)^(h/12) × terminal P/B. The terminal multiple is applied once; h=3, 6, or 12 months. Price change = conditional price / reference price in the same currency − 1.
Inputs
Exact instrument, share class, and currency; verified per-share basis and reporting period; current raw reference price. Growth and the terminal valuation multiple are the user’s explicit assumptions; they are stored separately from the source data.
Assumptions
EPS or BVPS changes at the specified constant annual rate, and the selected multiple may be applied at the end of the horizon. This is an algebraic sensitivity analysis. ETF and crypto receive a historical risk / Monte Carlo profile, not a company EPS price target.
Limitations
Not a price-target estimate or total-return forecast; excludes dividends, taxes and currency effects. No multiple-based result is produced for non-positive or unverified EPS/BVPS, stock-split uncertainty, stale prices or currency mismatches. Archived v1–v3 runs remain available with their own historical formulas.
Trend strength · ADX(14)
Formula
TR = max(H−L, |H−previous C|, |L−previous C|). +DM is the positive upward movement when it exceeds the downward movement; −DM is the reverse. Wilder smoothing: Sₜ = Sₜ₋₁ − Sₜ₋₁/14 + xₜ. DX = 100 × |+DI−−DI| / (+DI+−DI); ADX is the Wilder average of DX.
Inputs
Consecutive daily H/L/C values based on the same adjusted basis. The first ADX14 appears at zero-based index 27, after 28 valid candles. Initialization restarts when a candle is invalid.
Assumptions
ADX measures trend strength, not direction. With zero movement, DX is zero; there is no arbitrary micro-price threshold. Strict mode calculates only from validated H/L/C candles.
Limitations
Not a buy or sell signal and not an expected return. The former internal JEL variable is retained only in historical reports.
Portfolio concentration · HHI and effective number of positions
Formula
HHI = Σwᵢ², where wᵢ is the weight of the same economic asset, between 0 and 1. Effective number of positions = 1 / HHI. The largest position and Top 5 weights are the direct sums of the sorted economic exposures.
Inputs
Positions with positive market value that can be converted to a common reference currency. The same ticker across multiple accounts is consolidated as one economic exposure. An item with a missing price or missing currency conversion is not treated as zero.
Assumptions
The ticker and listing have been correctly identified. HHI measures concentration of weights; the effective number of positions is the more readable reciprocal of the same measure. Neither metric independently classifies the portfolio as suitable or unsuitable.
Limitations
Direct-position HHI does not capture underlying ETF overlap without dated holdings data. Separate share classes of the same issuer, or economically related assets, may remain separate without precise issuer identification. Diversification does not rule out losses.
Correlation and covariance · aligned to common dates
At least 31 common, ordered daily price observations for the included assets; prices in foreign currencies are converted using the reference-currency series for the same dates. Weights are normalized across usable assets, and coverage is shown separately.
Assumptions
Historical daily log returns and sample covariance describe co-movement over the period examined. Stock and ETF calculations are annualized using 252 observation days. Every matrix cell is based on the same common date range.
Limitations
Correlation is not causation and can change quickly when market regimes shift. No result is produced with too few common days, incomplete FX data or differing price adjustments. The estimate excludes liquidity, taxes, fees and execution effects.
Historical volatility · annualized standard deviation of daily log returns
Formula
σannual = √(wᵀΣw) × 100. For an individual asset, this is the sample standard deviation of daily log returns × √252 × 100; √365 may be used for separate analyses of continuous crypto-market series.
Inputs
Log returns calculated from prices with the same time basis that are positive and ordered, and for portfolios, a covariance matrix and economic weights aligned to common dates.
Assumptions
The standard deviation of past daily observations over the period examined is a comparison measure. Annualization uses time-proportional scaling of daily variance; it does not assume a normal return distribution or constant future volatility.
Limitations
Highly period-dependent, and does not by itself show the direction of losses or the full risk of extreme events. With missing common data, coverage decreases or the calculation stops.
Maximum drawdown · largest historical decline
Formula
DDₜ = Vₜ / max(V₀…Vₜ) − 1. Maximum drawdown = min(DDₜ). The portfolio curve starts at 100 by successively exponentiating the weighted sum of common-daily log returns.
Inputs
Chronological asset or portfolio value series. For portfolios, the same common-date and currency treatment applies as in the covariance calculation.
Assumptions
The metric describes the largest observed decline from peak to trough within the selected time window. It is shown as a negative percentage.
Limitations
It does not show the duration of the decline, the time to recovery or losses outside the period examined. The largest historical decline is not an upper bound for the future.
ETF look-through · direct and indirect exposure
Formula
Indirect exposureⱼ = Σ(ETF market value × holdingⱼ weight). Total economic exposureⱼ = direct valueⱼ + indirect valueⱼ. An unknown holdings percentage is shown separately as “Unknown ETF composition”.
Inputs
Precisely identified ETF listing, dated issuer holdings source, unique holding identifiers and percentage weights whose sum matches the reported coverage. TER is separate, dated product data.
Assumptions
The disclosed holdings and weights apply to the specified ETF on the indicated data date. A partial Top 10 or other extract covers only the percentage actually known and is not normalized to 100%.
Limitations
No look-through without a holdings source, data date or consistent weights. Trading since disclosure, derivatives, cash and multi-level fund-of-funds exposure may cause discrepancies. An ETF may contain multiple components, but this does not automatically make it diversified.