---
name: graham-deep-value
lens: Statistical Deep Value
attributed_to: Benjamin Graham
style: value
markets: Any market with reliable balance-sheet disclosure; strongest where a wide basket can be held
description: Buy a statistically cheap basket against verifiable balance-sheet value, demand a quantified margin of safety, and refuse to pay for forecasts.
---

# Statistical Deep Value Lens

> **Framework, not investment advice.** This is our own summary of a publicly
> documented methodology, reconstructed as a procedure. It is not written or
> endorsed by the person named, and any verdict it produces holds only under this
> framework.

## What this lens optimizes for

Not being wrong expensively. It replaces judgement about the future with
measurable, present, auditable value, then insists on paying meaningfully less than
that. It expects to be wrong on individual names and to be right on the basket.

## Market-fit note

- Works wherever balance sheets are audited and comparable, and where you can hold
  20+ names so idiosyncratic blowups average out.
- In A-shares, statistically cheap screens are dominated by financials, utilities
  and property; treat that concentration as a single macro bet, not diversification.
- In HK, persistent deep discounts often reflect a controlling-shareholder
  discount, not mispricing — see the disqualifiers.

## Priority signals (walk in this order)

1. **Solvency before cheapness.** Current ratio, total debt to net current assets,
   interest coverage, and whether the company can survive two bad years without
   raising capital. A cheap insolvent company is not cheap.
2. **Net current asset value.** Current assets minus *all* liabilities (including
   off-balance-sheet and lease obligations), against market cap. The classic
   threshold is a market cap below two-thirds of that figure — rare outside crises,
   and its rarity is information.
3. **Tangible book, adjusted.** Book value with intangibles and goodwill removed,
   then marked for what the assets would actually fetch: inventory obsolescence,
   receivable ageing, investment property carried at stale valuations, land carried
   at historic cost.
4. **Earnings stability, not earnings growth.** Positive earnings in each of the
   last 5–10 years. Averaged earnings across a full cycle, versus price. Growth
   forecasts are excluded from the valuation on purpose.
5. **Dividend record.** An uninterrupted payout history is evidence the reported
   earnings were real cash.
6. **Margin of safety, quantified.** Price versus the *conservative* value estimate,
   stated as a percentage discount. If you cannot state the number, the lens has not
   been run.

## Disqualifiers (hard vetoes)

- **Leverage that removes optionality** — debt maturing inside 12 months exceeding
  cash plus committed facilities. Cheapness is irrelevant against a refinancing wall.
- **Loss in any of the recent years under review** — the stability test is a gate,
  not a scored input.
- **Asset base that cannot be marked** — the "value" is goodwill, capitalized
  development cost, level-3 assets, or related-party receivables.
- **Value the minority shareholder cannot reach** — cash controlled by a parent,
  assets pledged to related parties, a controlling holder with a history of
  extraction. A discount that exists because value leaks is not a margin of safety.
- **Single-name concentration.** This lens is basket-shaped. Applying it to one
  position removes the statistical premise it depends on.
- **Value derived from a forecast.** If removing the growth assumption removes the
  discount, the case belongs to a different lens.

## Scoring rubric

| Gate | Pass condition |
| --- | --- |
| Solvency | Survives two years of zero growth with no external financing |
| Balance-sheet value | Discount to conservatively adjusted tangible value, stated in % |
| Earnings stability | No loss years in the review window |
| Distribution | Dividends paid, or a credible explanation for retention |
| Basket | ≥ 20 comparable names, or an explicit acknowledgement the premise is broken |

Verdict is the weakest gate, not the average.

## Typical misuse

- **Applied to asset-light businesses.** Software, brands and networks carry almost
  no markable book value; this lens will reject good businesses and reward bad
  balance sheets. Use a quality or GARP lens instead.
- **Applied to a single stock as a conviction case.** The methodology's expected
  return comes from a diversified basket absorbing individual failures.
- **Treating a value trap as a margin of safety.** Cheap assets in a structurally
  declining industry get cheaper: the denominator erodes faster than the discount
  closes. Stack a forensic-short or cycle lens to test this.
- **Ignoring the reason for cheapness.** The lens is deliberately indifferent to
  narrative, which becomes a defect when the narrative is "the assets are impaired
  and the auditor has not caught up yet".
- **Cross-border book comparison.** Accounting standards differ on revaluation,
  leases and capitalization; a discount to book across jurisdictions may be an
  artifact.

## Falsifiers

- The adjusted tangible value falls below the market cap once obsolete inventory,
  ageing receivables and stale property marks are applied → the discount was
  arithmetic, not real.
- A capital raise below book → management disagrees with your asset value.
- Repeated asset write-downs in consecutive periods → the book is a lagging
  indicator and the anchor is unreliable.

## Output contract

Report under the skill-level output contract. State the verdict as "under a
statistical deep-value framework, …", give the discount as a number with the
adjustments that produced it, and state explicitly whether the basket premise
holds. This is analysis, not investment advice.