Stock Forecast and Scorecard Methodology

Forecast snapshot

1,027 stocks have 3, 6, 9 and 12-month model forecasts based on September 21, 2026 data.

Inputs and feature categories

• Lagged stock returns, momentum, moving-average, volatility, drawdown, and prior-range measures.
• Company margins, valuation, free-cash-flow yield, revenue growth, and earnings growth.
• Broad-market and sector context, rates, volatility, macro conditions, and calendar-era context.
• Missing or inapplicable values remain explicit until train-fold-only processing; unavailable public values are never filled with a display placeholder.

Leakage and point-in-time controls

• Features use prior-session values or a documented publication lag. Fundamental observations activate only after their accepted or conservative availability date.
• Vintage-sensitive macro data keeps the release vintage available at the forecast origin rather than substituting a later revision.
• Each fit uses only the trailing five years of weekly training observations. Imputation and fitted transforms are learned inside the training fold.
• Validation is time ordered and walk-forward; random train/test splits and in-sample success claims are not used.
• A 252-session embargo approximation separates training and evaluation origins. Exact target labels still use the first verified market session on or after the 3, 6, 9, or 12-month calendar date.

Estimates, ranges, and calibration limits

• Lower Estimate, Median Estimate, and Upper Estimate are target-date model outputs when genuine range evidence is available.
• Each 3, 6, 9, and 12-month value is an independent target-date estimate. The points are not connected into a path or fan.
• The lower–upper estimate range is not a confidence score, guaranteed trading range, or prediction that prices will stay between two values.
• A median-only snapshot does not support an inferred lower–upper range. Crossed or otherwise invalid model endpoints fail closed rather than being sorted or repaired for display.
• Estimated price change is always measured from the forecast's stated starting price, not from today's price.

Model evaluation uses untouched out-of-sample observations. Diagnostics include quantile pinball loss, weighted interval score, empirical interval coverage and coverage error, interval width, tail rates, and raw quantile-crossing checks, as applicable.

Model evaluation

Independent model qualification combines data checks, time-ordered out-of-sample evaluation, weighted evaluation, and release review. Each forecast snapshot retains its own data date and starting price.

Track-record metrics require exact target-date observations to settle.

Numeric model and narrative separation

LightGBM produces numeric estimates. A language model may explain supplied data and already-produced estimates, but it cannot create, replace, recalculate, or qualify a numeric forecast. Third-party analyst targets are also separate from Vaniras model estimates.

How stock scores are calculated

All scores in this Scorecard describe relative characteristics as of the shown date, using historical market data and reported company information; higher scores on the 0–100 scale indicate more favorable characteristics in the relevant comparison group, and none forecasts future performance.

The Overview and Scorecard tabs show the same dated company snapshot. Scores are independent of the selected 3, 6, 9 or 12-month forecast horizon: selecting a horizon does not turn them into scores for that future period. Financial measures use reported trailing twelve-month (TTM) figures where applicable; balance-sheet measures use their shown reporting date. Reported periods can end before the score date. Safety and Momentum use recent trading history over roughly three and six months.

Downside Resilience: Downside Resilience combines the Safety and Quality readings: recent price volatility, the distance below a recent price high, cash flow and profitability, each compared with companies that have comparable scoring inputs. In historical research on the stocks covered here, higher readings coincided with fewer price declines of 25% or more over the following three to twelve months. It describes past downside behavior and current financial characteristics; it does not forecast returns, direction or the size of any future decline. The Model Forecast covers estimated price ranges. Downside Resilience is shown when Safety and Quality both have a score and a comparable peer group.

Earlier Overall: Overall brings together Value, Quality, Growth, Safety and Momentum to summarize the company’s financial and market characteristics. Overall is shown when all five factors have a score and a comparable peer group. Each factor is assessed independently; a factor can be shown without an Overall score or a price forecast. A higher score indicates a stronger combined profile among companies with comparable scoring inputs. Strong individual factors can offset weaker ones, so review the breakdown alongside the total. Cash generation and Balance strength are shown separately and do not contribute to Overall.

• Value: Value compares the company’s market valuation with its reported earnings and cash generation. For financial companies, balance-sheet equity is used where it provides a more suitable comparison. A higher score indicates more favorable valuation measures within the relevant group. Losses remain part of the assessment. A high Value score does not establish that a stock is undervalued or explain why its valuation is low.
• Quality: Quality compares reported profitability, including how much profit the company earns from its revenue. Financial companies use profitability measures suited to their business structure. A higher score indicates stronger profitability within the relevant comparison group. This score does not independently assess accounting reliability, earnings consistency or the durability of a competitive advantage.
• Growth: Growth compares changes in reported revenue and profit between the latest trailing twelve-month period and the preceding twelve-month period. It accommodates both profitable and loss-making companies, including transitions between profit and loss. A higher score indicates stronger reported growth within the relevant group. A single strong comparison does not establish a sustained growth record.
• Safety: Safety combines recent price volatility, the distance below a recent price high, and selected cash-flow and profitability measures. A higher score indicates a more favorable combination of these characteristics within the relevant group. It is not a credit rating or a comprehensive assessment of financial distress, and it does not measure the largest historical peak-to-trough loss.
• Momentum: Momentum compares price performance over recent trading periods, including performance relative to a broad-market benchmark. A higher score indicates a stronger recent price trend within the relevant group. This benchmark comparison does not fully adjust for the stock’s market sensitivity. Momentum measures price behavior rather than the company’s financial quality or valuation.

Additional company measures

• Cash generation: Cash generation compares reported free cash flow—the cash remaining after capital expenditure—with reported revenue. A higher score indicates stronger cash generation per unit of sales among eligible companies in the same sector. Negative free cash flow remains negative in the comparison. Investment timing and working-capital movements can affect the reading. This measure is shown separately and does not contribute to the headline score.
• Balance strength: Balance strength compares cash and equivalents with consolidated debt, relative to the company’s asset base. A higher score indicates a more favorable cash-and-debt position among eligible companies in the same sector. This is a focused balance-sheet measure: it does not assess debt maturities, borrowing conditions or the company’s full repayment capacity. This measure is shown separately and does not contribute to the headline score.

These additional measures compare eligible companies in the same sector with aligned reporting inputs. Cash generation uses the shown TTM period; Balance strength uses the shown balance-sheet date. They are not shown for financial companies or real estate, whose business structures need different comparisons.

Sector Avg. comparison

Sector Avg. is the arithmetic mean factor score of eligible Vaniras-covered companies in the same sector and scoring snapshot, using matching calculation inputs. The subject company and its other share classes are excluded; each peer issuer contributes once in the issued universe order. At least five eligible peers are required for each factor. A factor without sufficient comparable peers shows N/A; dates and score versions are never mixed. The comparison shows the existing scores against a sector benchmark; it does not recompute Overall or turn the main factors into sector-adjusted scores. Stock-minus-sector differences are score points, not percentage returns.

Comparison groups and interpretation

Peer positions and data coverage can differ by factor. Current stockwide snapshots compare matching available inputs; Value and Quality also account for differences in financial reporting. These groups do not fully adjust the main factors for industry or company size, and their membership can change between snapshots. A score of 80 is not an 80% chance of a price increase, and a score change can reflect changes in peers as well as in the company. Review the shown score date, peer count and coverage alongside the reading.

Historical and other existing score views: The factor descriptions above apply to the current research display in Overview and Scorecard. Older snapshots and other existing score views retain their issued definitions: Value uses positive P/E and cash-flow comparisons, Quality uses gross and operating profitability, and Growth uses revenue and earnings-per-share growth. Older snapshots may also use a smaller fixed peer group. Their question-mark explanations reflect those definitions; avoid treating different methods or comparison groups as one continuous score history.

Sources

Limitations and risk

• Markets, companies, regimes, corporate actions, and source data can change in ways historical training does not capture.
• Missing point-in-time history can exclude a stock from numeric estimates even when it remains visible in the product list.
• Descriptive scores are not price forecasts or recommendations. Analyst targets can use different methods and dates.
• Stale snapshots stay tied to their original starting price, data date, and target dates.

Vaniras provides model-based estimates, not personalized investment advice or trade instructions. Forecasts use historical and market data and can differ materially from actual outcomes. Analyst targets, when shown, are third-party estimates and are separate from the Vaniras model.