Independent learning lab · framework lineage

Morgan Stanley / 摩根士丹利：base-rate and reference-class lab

Use historical counts to study event rates and a Wilson 95% interval, then use observations to calculate mean, median, sample standard deviation and a descriptive z-score. The lab checks structural probability-disclosure gates but does not replace independent verification of the sample list, freeze time and sources.

Source document

Counterpoint Global Insights — Bayes and Base Rates 2.0

PDF physical pages p1–6 contain the methodology; p7–10 are notes / disclaimer

Concordal is an independent educational tool and is not affiliated with, partnered with, sponsored by or endorsed by Morgan Stanley. The name identifies public framework lineage only; this page uses no institutional logo or proprietary model and gives no buy/sell advice.

Program 01 · start with the outside view

Reference-class event rate and 95% interval

Define the comparable sample and event before asking what usually happens. The page uses a Wilson score interval, avoiding the instability of a simple Wald interval in small samples or extreme rates.

p̂ = k ÷ n · Wilson 95% CI = (p̂ + z²/2n ± z√[p̂(1−p̂)/n + z²/4n²]) ÷ (1 + z²/n)

Outside-view and reference-class lineage: PDF physical pages p1–6. The Wilson interval and n≥30 / leakage-free gates are Concordal’s transparent implementation, not formulas reproduced from the PDF.

Reference-class inputs and audit declarations

Single active reference class

State market, business model, scale, profitability stage, region, time window and event definition; “peers” alone is insufficient.

Independent observations ninteger; probability gate requires ≥30

Observations meeting the event definition kinteger, 0 ≤ k ≤ n

There is one active reference class; it was not switched after seeing the result.The sample and event definition were frozen before outcomes, with no future information or ex-post survivor screen.Observation windows do not overlap and one company or event is not repeated as multiple independent samples.

FORMULA PASS · independent sample evidence still required

This is an input-quality gate, not a claim that the event will or will not occur.

One non-empty, currently active reference class is defined
n and k are valid integers with 0 ≤ k ≤ n
Sample size n = 40; gate requires n ≥ 30
The 95% interval satisfies 0 ≤ lower ≤ event rate ≤ upper ≤ 100%
No future leakage or ex-post survivor screen
Observation windows are non-overlapping and independently counted

Descriptive event rate k/n

27.50%

The teaching formula gate passed; only a preserved and independently verified sample list, freeze time and sources can support real research disclosure

95% interval lower bound

16.11%

95% interval upper bound

42.83%

Program 02 · inspect a distribution, not one point

Mean, median, sample standard deviation and descriptive z-score

Enter historical observations and one forecast. The z-score answers only “how many sample standard deviations is the forecast from the sample mean”; this lab never converts it into a normal-tail probability.

s = √[Σ(xᵢ − x̄)² ÷ (n − 1)] · z = (forecast − x̄) ÷ s

Distribution-distance and z-score lineage: PDF physical page p4; Concordal explicitly supplies the n−1 sample-standard-deviation and calculator implementation.

Historical distribution inputs

Historical observations

Separate values with commas, semicolons, spaces or new lines; all values must use one unit, such as revenue-growth percentage points.

Forecast to comparesame unit as observations

Valid observations

Used for descriptive statistics only

Mean

13.75

Median

13.50

Sample standard deviation s

3.57

Denominator is n−1

Historical range

8.00 – 21.00

Descriptive z-score

1.75σ

Distance from the sample mean only; never converted into event probability

z-score is not a probability

Real company growth, return and valuation distributions may be skewed, fat-tailed, segmented or regime-dependent. Without testing the distributional assumptions, a normal-tail area must not be presented as the “probability the forecast occurs.”

How to use this lab correctly

• Start with a broad reference class and treat industry, scale or stage filters as sensitivities; narrower classes easily reduce both sample size and transferability.
• Separate nominal from real values and organic from acquired growth, and show time slices across market, rate or technology regimes.
• The page cannot verify your leakage-free or non-overlap declarations; real research must retain the sample list, freeze time and source evidence.

Back: Goldman value-creation lab