Technical specification

What is specified

Each area below is specified in full unless the right-hand column names something still to be fixed. A rule that is not yet fixed will be published here before any live figure depends on it.

π rating

Zone lines

These are PIInneh’s classification lines, used by every page, report and index.

• Altman Z (operating companies): safe above 2.99; grey from 1.2 up to and including 2.99; distress below 1.2.
• Merton distance to default (financial companies): safe at 4 standard deviations or more; grey from 1.2 up to 4; distress below 1.2.
• Cash runway, DD_Bio (biotechnology companies): safe at 18 months or more; grey from 6 up to 18 months; distress under 6 months.

How a rating moves

A rating moves when its model crosses a zone line. Crossing from safe into grey takes it down one notch; falling from safe straight into distress, two notches. Rising from distress into grey takes it up one notch; rising from distress straight to safe, two. Crossing between grey and its neighbouring zone moves it one notch. The previous rating is always kept, so every move is visible.

In a Black Swan scenario a rating is cut by one notch for each zone its model falls through, and it is never raised. Biotechnology ratings move with the cash runway and the results of trial readouts.

Companion models

Beneish M, Piotroski F, Ohlson O and Zmijewski are shown beside the rating of operating companies. Until the rating specification is published with the first live ratings, they do not change the rating. None of the corporate models is ever applied to financial or biotechnology companies.

Signals

Directions and agreement

Each of the five method groups (forensic ratings, the Superradiance trend, estimates, price against intrinsic value, and insider and activist activity) gives each stock one direction: up or down. The estimates group takes its direction from the annual estimate revisions; the quarterly results are shown beside them. A stock’s agreement is the number of other groups pointing the same way. The Strongest Agreement list shows the stocks on which at least three groups agree, ordered by the number agreeing, then by the share agreeing, then by ticker, and shows the top 15. When a report needs one overall direction and the groups are evenly split, it counts as up.

Screens

• Value trap: the price is below intrinsic value while the forensic rating or the trend points down.
• Quality at a discount: the price is below intrinsic value and the rating was upgraded.
• Smart money: insiders are accumulating while the trend points down.
• Distress watch: a Merton probability of default of 5% or more, or a Zmijewski score above 0.

Checking a call

Every call is stored on the day it is issued and checked against the daily close 20 trading days later and 63 trading days later (about one and three months). It is right when the close has moved in the direction it pointed; losses are included, not dropped. Four calls on a stock give its record in steps of 25%, so a single stock’s record is an indication, not a precise accuracy.

A report also seals the breaking level it names on the day it is issued. That call has broken if any daily close in the next 63 trading days reaches the level; until the 63 days are over it is shown as held so far, with the number of days already checked, and after them as held.

Value and estimates

Discount and margin of safety

The discount or premium is the gap between intrinsic value and the price, measured as a share of the price. The margin of safety is the same gap measured as a share of the intrinsic value, so it says how much of the value would have to disappear before the price was no longer covered. It is shown only when the price is below the value.

The growth the price implies

PIInneh asks what yearly earnings growth over the next ten years the price is already assuming. It values the company’s future earnings, growing at a trial rate for ten years and at 3% a year after that, and discounts them at a required return of 9% a year; it then adjusts the trial rate until that value equals today’s price. The starting earnings are PIInneh’s base estimate for the current fiscal year, or the analyst consensus where PIInneh has none.

The search runs from a 50% yearly fall to a 100% yearly rise; a price that needs more than that is shown as needing less than −50% or more than 100%. The figure is not calculated when the starting earnings are zero, negative or missing, nor for biotechnology companies, whose prices reflect their pipelines rather than current earnings. The answer depends on the 9% and the 3%: a higher required return or lower long-term growth raises the growth the price implies. It treats all of a company’s earnings as cash available to its owners, so it leaves out the investment needed to grow; for a company that reinvests heavily, the growth the price implies is understated.

Still to be published

The valuation model used for each type of company (discounted cash flow, residual income, earnings power or sum of the parts), and its inputs, will be published with live values.

Price projection

The inputs

144 technical indicators in eight families: trend (24, such as moving-average spreads, MACD, ADX, Aroon, Ichimoku and Supertrend), momentum (24, such as RSI, rate of change, stochastics, CCI and KST), mean reversion (18, such as Bollinger, VWAP and Keltner positions and robust z-scores), volatility (18, such as Donchian position and breaks, squeezes and ATR expansion), volume and flow (24, such as OBV, Chaikin money flow, MFI and cumulative volume delta), market structure (14, such as channels, support and resistance room and points of control), microstructure (12, such as close location, wicks and Amihud illiquidity) and statistical context (10, such as Kaufman efficiency, autocorrelation and the Hurst exponent). Alongside them the model reads 25 factors from the Factor Zoo, spanning quant, fundamental and macro data, computed in PIInneh’s backend. In this preview edition the chart runs on a simulated price feed and reads the 144 indicators only; the backend factors join it when live data is connected. Each indicator’s exact formula and window will be published with live projections.

How the indicators are weighted

Each indicator is turned into a reading between fully down and fully up. The model estimates the market’s condition as chances across six regimes: a persistent trend, a sideways range, volatility compression, a breakout, liquidity stress and event uncertainty. It replays the last 400 bars, issuing a test forecast every 5 bars for every horizon, and scores each test only once its outcome is known, so it never learns from information it would not have had at the time.

Each indicator and factor earns weight by how well it has predicted this stock, at this horizon and in each regime, with recent results counting more. Each family receives a share of the weight from the record of its best members and its fit to the current regime, and indicators that simply repeat their family are weighted down, to as little as about half of their weight. Direction is decided by the weighted evidence, never by a count of indicators.

The forecast and how it is checked

Forecasts can be made for 12 to 63 bars ahead. Each shows a central path with an 80% and a 95% range; the 80% and 95% are targets. The ranges come from the model’s own past errors on this stock and timeframe, and a final adjustment is then applied to the forecast shown. The coverage the chart reports is checked end to end: past forecasts, one every 12 bars over the last 300 bars, are issued by the same calculation as the forecast shown, with the same rolling 400-bar training window and only the data available on their date, and their final 80% and 95% ranges are scored against the closes that followed. A horizon needs at least six scored forecasts. The check runs in the background after the chart opens, and the coverage reads “checking…” until it is done. The model’s own reliability score uses a faster check inside its training replay. The coverage shows how closely the targets have been met, not a guarantee that they will be. A move counts as up or down only when it would clear the estimated cost of trading. Reliability, from 22% to 95%, falls when the indicators disagree, the regime is unclear or few tests have matured. In locked mode each forecast is frozen when it is issued and marked as a hit or a miss when it matures, and the last 24 are kept for the session; rolling mode reissues the forecast at every bar, so its forecasts are not kept.

Thesis break prices

Altman Z

Only one part of Altman Z moves with the share price: the market value of the company’s equity against its liabilities. PIInneh holds the other parts as they are and finds the share price at which that part has changed enough to bring Z to the condition’s line. The market-value part is always a positive share of the size of Z, so a lower share price always lowers Z, including for a company whose Z is already below zero.

Merton distance to default

A change in the share price changes the value of the company’s assets by the equity’s share of those assets. The distance to default changes by that change in asset value, measured in units of the assets’ volatility. PIInneh finds the share price at which the distance reaches the condition’s line. A probability-of-default line is first turned into the distance to default that gives that probability. The probability is the standard normal model’s reading of the distance: a model assumption, not an observed default rate. PIInneh’s zone lines for the distance (4 and 1.2 standard deviations) are its own classification choices. In this preview the distances are sample data; how they are built (the default point, the horizon, and how asset value and volatility are estimated) will be published with live data.

Cash runway and catalysts

A cash-runway condition breaks on the date the months of cash left reach its line, counting 30.4 days to a month. A catalyst condition breaks on its readout date.

Status and edge cases

When every condition passes, a thesis is Intact; when one fails, Weakening; when two fail or the rating is cut below the rating it had when you added it, Broken. When no share price can reach a line, the condition shows “not reachable by the price alone”. A condition whose measure is missing counts as failed, and so does a value condition when there is no intrinsic value. The nearest break is the passing price-driven condition closest to the price, in percent.

Break chances and calibration

The chance of a break

A break is a daily close at or beyond the breaking level. A stock’s volatility is the standard deviation of its last 60 daily log returns (each close against the one before, in logarithms), from its last 61 closes; at least 20 valid daily changes are needed, and with fewer no chance is given (“not enough price history”), never a chance of zero. PIInneh gives the chance that a stock moving with that volatility closes at or beyond the level at least once within the period: one month is 21 trading days and three months 63. It assumes no drift in the logarithm of the price, which means a small upward drift in the price itself (about +2.9% over three months for a stock with 3% daily volatility). Because only daily closes are checked, not every moment of the day, the chance is corrected downward with a standard adjustment for checking at intervals (Broadie, Glasserman and Kou); the adjustment is an approximation, least exact when the level is very close to the price.

A condition that has already failed shows “already failed” instead of a chance. A level that no price can reach shows a dash. A stock whose price has not moved at all gets a chance of 0.

Earnings day

Separately, PIInneh gives the chance that the earnings-day move alone reaches the level, using the stock’s typical earnings-day move, and flags a condition as able to break on that day when its distance from the price is within that typical move.

The calibration test

From the close 63 trading days ago, PIInneh takes the chance it would then have given of a close reaching levels 3%, 6%, 10% and 15% above and below the price, and checks whether a close in the following 63 trading days did. The results are grouped by the chance given (0–10%, 10–25%, 25–50%, 50–75% and 75–100%). A group is well calibrated when the share that actually broke is within 10 percentage points of the average chance given. Groups with no results are not shown.

Correlation and diversification

Correlation

The correlation of two stocks is the standard (Pearson) correlation of their daily returns over the last 252 trading days, about one year. Only the days on which both stocks have a return count for the pair, for the covariance and for both stocks’ volatilities alike. A pair needs at least 20 shared days: with fewer, its correlation is not available, never zero. Every result is kept between −1 and 1. The 450-company correlation table of the QuantGDI indices is recalculated weekly.

Market Link

A company’s Market Link is its average strength of connection with the other companies of the 5,000-company universe, counting a negative correlation as strongly as a positive one. A low Market Link means a company moves largely on its own.

Diversification

On a watchlist, two stocks with a correlation of 0.75 or more are flagged as moving so closely that holding both adds little, and PIInneh suggests a replacement in the same industry. The number of truly independent stocks a list behaves like falls as its average correlation rises: ten stocks with an average correlation of 0.5 spread risk about as well as two independent stocks. This is an approximation that assumes equal positions and similar volatilities.

In this preview

The correlations come from a sector model (the two companies’ connection with the market and the connection between their sectors, plus a pair-specific part), not from price history.

Index rules

• Universe: 5,000 companies, 250 in each of 20 industries.
• QuantGDI 450: the 450 companies with the lowest Market Link.
• QuantGDI 50 and 150: built one company at a time from the QuantGDI 450, starting with the lowest Market Link. Each step adds the company with the lowest average correlation with the companies already chosen, counting negative correlations as zero; ties go to the lower Market Link, then to the ticker. Because the steps are the same, QuantGDI 50 is the first 50 of QuantGDI 150.
• The other families are drawn from the QuantGDI 450. Their selection rules will be published with live index values; in this preview their members are chosen by a sample score.
• Weights: Active Floating Market Cap Weighted + 10% Cap. Each company starts with its value as a share of the index total. Any company above 10% is set to 10%, and the excess is shared among the other companies in proportion to their weights; this is repeated until no company is above 10%. There is no minimum weight. Until float data is connected the value is the full market capitalization; the active-float definition and its data source will be published then.
• Calendar: reconstituted every quarter, at the end of March, June, September and December, effective at the next trading session. The calendar starts with the first-quarter 2025 reconstitution (March 31, 2025), and the history before it is a back-test; in this preview edition every index level, before and after that date, is a sample series.
• Halts: a halted company keeps its last price. When fewer than 45 of the 50 QuantGDI 50 companies have a price, the index carries a halt notice.
• Levels: in this preview the index levels are sample series. How levels are calculated from constituents (price or total return, the divisor and corporate actions) will be published with live index values.

Black Swan calculations

The scenarios

Each scenario is a set of shocks to twelve market factors. A negative number is a fall in that factor; volatility and liquidity stress rise:

Watchlists

A stock’s move is the sum, over the ten market factors, of its industry’s sensitivity to the factor times the factor’s shock, with the market counted in full and the other nine at half. It is then scaled by a company factor between 0.85 and 1.15. A fall is made deeper for lower-rated companies: for a πC company by a fifth of the scenario’s liquidity stress (with liquidity stress of +0.90, a fall 18% deeper), for each higher grade proportionally less, and not at all for πAAA. No stock falls more than 90%.

Each industry’s sensitivity to each factor:

The thesis is then re-checked at the shocked price. Altman Z is also lowered by one and a half times any earnings shock (an earnings shock of −20% lowers it by a further 30%). Merton distance to default shrinks as volatility rises (a volatility shock of +100% cuts it by a third). The cash runway is cut by credit and liquidity stress, to no less than 40% of its months, on the assumption that no new money is raised. Earnings estimates count as cut when the earnings shock is −8% or worse.

Indices

Each company takes the shocks through its sector’s sensitivity to each factor, adjusted for the company by a factor between 0.85 and 1.15. The sector sensitivities:

The shocks are combined so that losses compound, so no holding can lose more than everything; each company’s result stays between a 95% fall and a 200% rise. A company’s quality cushions its fall, weighted 30% quality, 30% balance sheet, 25% earnings and 15% liquidity; its fragility deepens it, weighted 18% crowding, 25% volatility, 16% illiquidity and 16% weak earnings. Volatility rises with the scenario’s volatility shock.

Each holding’s selling pressure, from 0 to 1, adds 1.40 for each unit of its fall, 0.45 for each unit of its stressed volatility, 0.25 for each unit of illiquidity, 0.20 for each unit of crowding and 0.25 for each unit of weak earnings. Under stress each holding’s weight shrinks the more its pressure, at the scenario’s selling-pressure rate, and the weights are then made to add up to 100% again. A sector’s pressure adds up its holdings’ falls, their liquidity drag and 0.65 for each unit of weighted pressure.

Trading those weight changes costs, for each holding: half its stressed bid–ask spread, plus price impact (the scenario’s price-impact rate, times the stock’s stressed daily volatility, times the square root of the trade’s share of the stressed daily traded value), plus a spread-stress charge (the scenario’s spread-stress rate, times half the spread, times the spread’s size against daily volatility, capped at 5). Stressed daily traded value is the normal traded value cut by the scenario’s traded-value cut, times liquidity stress, times the stock’s illiquidity, but never below the scenario’s floor; the stressed spread is the normal spread widened by the spread rates for liquidity stress and for the volatility shock. The liquidity drag is the sum of each weight change times its cost. In this preview the spreads, traded values, quality, liquidity and crowding of each company are model assumptions keyed on its sector and size.

The index’s return is the weighted sum of its companies’ returns. Its volatility combines theirs with the correlation the scenario assumes between members, which is set for indices of 50, 150 and 450 companies and read in between for other sizes. Under stress, weight moves away from the holdings under most selling pressure, and trading that change costs half the stressed bid–ask spread plus the price impact of the trade against the stock’s stressed daily volume, at your account size: this is the liquidity drag.

The central fall is the index’s return after the liquidity drag and an allowance for its volatility. The range around it widens with the volatility and the liquidity drag; its low end is never worse than a 99% fall and its high end never above no fall. Recovery time is how long the index would take to climb back from the central fall at its expected yearly growth after the stress: its members’ expected return, plus 0.03 for each unit of quality, 0.03 for balance sheet and 0.02 for liquidity, less 0.025 for each unit of the scenario’s recovery drag, and never less than 2% a year; it is at most ten years (2,520 trading days). The resilience score, from 0 to 100, is an S-shaped curve of one combined measure: the central fall counts against it with a weight of 2.40, the liquidity drag 8, each year of recovery 0.35 and concentration 0.50, and the quality cushion counts for it with a weight of 0.45. The curve’s midpoint is −0.65 for severe scenarios, −0.55 for historical replays and −0.45 for moderate ones, and its steepness is 8. Scores fall together in harsher scenarios, so read them as a ranking.

The settings of each scenario

How strongly each scenario applies each part of the calculation above (its liquidity settings follow in the next table): the quality cushion and the fragility penalty, how far volatility rises, the correlation between members under stress, the selling pressure on weak holdings, the volatility allowance in the central fall, how wide the range is made by volatility and by liquidity, and how much the stress slows recovery.

The liquidity settings of each scenario:

Versions

The current methodology is version 1.2, effective October 16, 2026, computed by engine build 4c5025004b43 on sample data. Every report names, in its header, the methodology version and the engine build it was computed with, so it can be traced to these rules and that code. Each version and what it changed: