Valuations ++ v6.x

The chart now states whose numbers it is showing.

The bear / base / bull levels this site computes from management’s own filed guidance and each company’s own valuation history now travel to the TradingView chart itself — Pine cannot fetch a URL, so the levels ship inside the indicator as a data library, dated like any other snapshot. The scenario table then orders its evidence: filed guidance first, street targets second, the model third — and the header names the tier that is speaking. A name the library does not cover draws nothing and says so. Never a silent substitution.

The v6.5 scenario-table header and its two new rows (MTARTECH, 02-Aug-2026). The 🏛️ Filed row is this site’s own range at its target FY — deliberately not squeezed into the 1/2/3-year grid — with the exit multiple, the guided rate and the snapshot date in the tooltip. The 🎯 Street row carries the analyst low / average / high with the estimate count, because on NSE mid-caps that count is often one.

Covered names also draw the three levels as lines in the panel, each labelled the way a level should be: 🏛️ BASE 6562 · FY27 · 87.2x — the price, the year it lands on, and the multiple behind it. All of it arrives as plain constants: zero of the indicator’s data-request budget spent.

Arithmetic on management’s own filed guidance and each company’s own past valuations — not price targets, not a recommendation.

The band now knows whether management keeps its word.

v6 added the guidance layer: what management guided each quarter, what was delivered, and a running delivery score per company. v6.2 goes further — where usable guidance exists, the fair-value bands themselves re-base on an evidence-weighted blend of the analyst estimate and management’s own number (backtested identity weight × credibility × call-date freshness, capped at 45%), and the scenario engine moves with it. No guidance, toggle off, or a TTM anchor — the analyst-only basis returns exactly. 641 covered names carry a usable figure today; every one discloses its blend on the chart.

The estimates panel as it renders on the chart (MTARTECH, 01-Aug-2026) — hover the Guidance cell: management’s own words, the bridged EPS figure, the exact blend applied, the disagreement warning and the source filing, exactly as on TradingView.

Descriptive analytics on public disclosures — not a recommendation. Scores show evidence, never a signal.

Every valuation model has the same weakness.

It picks a multiple — usually an average or a median — multiplies it by an earnings figure, and prints one number. That number carries no sense of how reliable the multiple was, whether the company still deserves it, or how wide the plausible range actually is.

Where does the price sit?

Drag the price. The band is fixed — it comes from the company's own multiple history — and the verdict simply reports where price falls inside it.

Illustrative band. The three margin-of-safety rungs sit at 10, 20 and 30% below fair — turning “wait for a better price” into a specific number.

Nothing gets silently dropped.

Hard cutoffs are the usual approach — discard any P/E over 500, any P/S over 100. The effect is that genuinely high-multiple companies are never sampled at all, their history stays empty, and the model quietly falls back to something derived from today's price. It then reports “fairly valued” for exactly the names where valuation matters most.

Instead, incoming samples are winsorized at the rolling 1st and 99th percentiles, with a 25% headroom envelope. Genuine re-ratings creep in; a one-day data spike is capped.

Almost none of this is hand-set.

The reason a valuation model usually cannot be trusted is that someone chose the inputs, and those choices are invisible by the time you read the answer. Here the inputs are derived from the instrument on your chart, per company, and they change on their own when the company does.

Change the symbol and everything re-derives. A bank gets P/B-anchored weights with EV/EBITDA at zero. A loss-maker drops P/E entirely. A company whose multiple has structurally re-rated has that method removed from its own blend. None of that requires you to notice, or to know the sector, or to remember to change a setting.

The honest limit of that: automatic is not the same as correct. Industry detection reads the exchange’s own classification and will mislabel an oddly-categorised company; the classifier can only see what the financial data reports. Everything it decides is shown on the panel precisely so you can overrule it.

One multiple is a guess. Four, weighted, is a method.

This is not a stylistic preference — it is the finding of the accounting-research literature, and it is why the model refuses to hand you a single number.

Industry decides which multiple can even apply.

Before the coefficient-of-variation blender runs, the industry map sets the starting weights — because for some businesses a multiple is not merely less useful, it is meaningless.

So the weighting has three layers. The industry map sets a prior on what can apply at all. The company classifier adjusts it for what kind of business this actually is. Then the coefficient of variation refines it by how consistent each multiple has been for this company — and the stationarity gate removes any whose history has stopped being evidence.

One honest caveat on the research: Liu, Nissim and Thomas measured accuracy against contemporaneous market prices, so it is a test of explaining what the market already does — not proof of predicting what it will do next.

When a company's own past stops being evidence.

A business that has structurally re-rated has a multiple history spanning two incompatible regimes. Averaging across that break produces a fair value describing a company that no longer exists.

Each multiple's history is split in half and the two means compared. If they differ by more than 0.75σ, that method is dropped from the composite entirely.

Three futures, with probabilities attached.

Beyond the band, the script projects bear, base and bull paths across one, two and three years from a growth corridor bounded by the industry's own floor and ceiling — then attaches a probability to each and reports an expected value.

Since v6.5 the table also orders its evidence. Where this site has computed a filed-guidance range for the company, those levels lead and the table says so; where the street covers the name, its target range appears with the analyst count; the engine's corridor is the always-on floor beneath both. The firing tier is stated in the header — never silently substituted.

Probabilities illustrate the output shape. They are not a forecast for any particular company.

Two settings that matter more than all the rest.

What to look at, in order.

What this cannot do.

Run it on your own charts.

Valuations ++ is invite-only on TradingView. It pairs with IN Analytics Pro 3, which can consume this fair value directly through a source bridge.