MantaRisk Exposure exhibit Live MantaCore output 22 Jun 2026

Labels describe one portfolio. The factor model finds another.

Too many family offices still analyse and report on investments through reference-data classifications such as domicile, asset class and GICS sector. These views, which were once the norm, are no longer preferred by institutions as they create a misleading impression of the portfolio’s characteristics.

Here we take the example of a USD 9.8 million portfolio. A conventional reference-data breakdown presents it as broadly diversified. A returns-based factor model instead measures the economic drivers behind each holding and shows a much different picture of where risk and exposures actually reside.

The divergence is material at almost every level. By setting the two perspectives side by side, we illustrate why family offices should update their analytics to better serve their clients.

$9.8M

Portfolio value

~45%

Vol from 2 names (13% wt)

0.58

R² — factor-explained

~5%

Vol from the 28% bond sleeve

Weight says diversified. Risk says concentrated.

same 14 holdings · one shared 0–23% scale

Reference data · portfolio weight

How it looks

Market-value weight per holding. Nothing above ~10%; the bonds are the biggest lines. A textbook balanced allocation.

Austria 211710.3%

US Treasury10.2%

Novo Nordisk7.5%

PEMEX7.1%

Unilever7.0%

Novartis6.9%

Coinbase6.7%

Agnico / Newmont6.7%

Alibaba6.6%

TSMC6.5%

MicroStrategy6.3%

ASML6.3%

USD/NOK5.1%

On the shared scale the weights sit low and even — nothing dominant.

MantaRisk · contribution to volatility

Where the risk actually is

Each holding’s share of portfolio volatility. Two names dominate; the big bond lines all but vanish.

MicroStrategy22.8%

Coinbase22.6%

Alibaba9.3%

ASML9.3%

TSMC7.6%

Agnico6.7%

Newmont6.6%

Novo Nordisk6.0%

Austria 21172.8%

Unilever2.3%

Novartis2.2%

US Treasury1.1%

PEMEX0.8%

MicroStrategy + Coinbase = 45.4% of volatility on 13% of weight.

Reference data spreads the book evenly across fourteen lines and calls it diversified. MantaRisk’s decomposition shows two crypto-linked names carry nearly half the risk — the same two the factor model can’t explain at all.

The standard breakdowns disagree too

reference label vs factor β

By geography

Reference · domicile weightFactor · geographic β (%)

United States43.0%none

Austria10.3%none

Denmark7.5%9.3%

Mexico7.1%none

United Kingdom7.0%none

Switzerland6.9%2.7%

Canada6.7%none

Netherlands6.3%7.7%

Both columns share one 0–43% scale, so bar lengths are directly comparable. Reference = share of market value (sums to 100%); factor = regression beta as a percent (β×100, as in the platform). They measure different things — value share vs return sensitivity — so they won’t match one-for-one; that gap is the point.

Reference data spreads the book over eight countries, 43% of it “United States.” The factor model assigns a country exposure to only three — Denmark, Netherlands, Switzerland: the European operating companies (Novo Nordisk, ASML, Novartis). The US, Austrian, Mexican, British and Canadian lines carry no country factor at all — the US names load on sector factors (the model uses US/global as its base) and the bonds on rates. Domicile is a legal address; it is not the country whose market moves the holding.

By sector · equity sleeve

Reference · GICS weightFactor · sector β (%)

Info Technology28.5%8.0%

Health Care21.5%4.1%

Materials19.9%12.0%

Consumer Staples10.4%6.4%

Financials9.9%none

Consumer Discr.9.8%6.2%

Both columns share one 0–28% scale. Reference = share of the equity sleeve; factor = sector beta as a percent (β×100).

By label, Information Technology (28%) and Health Care (21%) are the biggest sector bets. By factor exposure the ranking inverts: Materials is the largest, “IT” shrinks — MicroStrategy, a third of the IT weight, loads on no factor at all — and “Financials” disappears entirely (Coinbase is unexplained). Health Care fades too, because Novo Nordisk loads on its country (Denmark), not the sector. Only Materials lines up. A GICS tag and the sector that drives returns are different things.

What the engine actually shows

risk ≠ weight ≠ beta

FINDING 01 · CONCENTRATION

Half the risk, two names, unexplained

MicroStrategy and Coinbase are 13% of the weight but ~45% of the volatility. Both come back with R² = 0 — the factor model finds nothing to attribute their moves to. Reference data calls them "Information Technology" and "Financials," implying they diversify the equity sleeve. They are its dominant, idiosyncratic risk.

MSTR 22.8% + COIN 22.6% vol · R² 0.00

FINDING 02 · HIDDEN BETA

The "safe" bonds: low vol, big duration

The 28% bond sleeve contributes only ~5% of volatility — so on a day-to-day basis the "AA+ government" label is fair. But it carries the portfolio’s largest single factor beta: the Austrian century bond carries a 201% beta to long government rates. That’s a concentrated rate-shock / scenario risk the volatility number understates and the credit rating hides entirely.

Bond vol 4.7% · Austria rates β 201%

FINDING 03 · LABELS

Country label ≠ country factor

Reference puts 43% in "US." The model expresses those names through sector factors (US is its base), and recovers genuine country factors only for the European ADRs — Denmark, Netherlands, Switzerland. A nuance, not a smoking gun: the exposure is real, just bucketed by sector rather than flag.

Geography factors: DK, NL, CH only

Factor betas — sensitivity, not risk

MantaCore LASSO · normalized

Portfolio betas, normalized to a fully-invested book and expressed as percentages (β×100, as the platform shows them). Note the inversion: Developed Government Bonds is the largest exposure (20.7%, pure duration) yet near-zero volatility contribution, while the crypto names that dominate risk show no beta at all. Beta tells you what a holding tracks; it does not tell you how much risk it adds.

Developed Govt Bonds · rates20.7%

Materials · gold miners12.0%

Denmark · geography9.3%

US Treasuries · rates8.9%

Technology · sector8.0%

Netherlands · geography7.7%

Consumer Staples · sector6.4%

Consumer Discretionary · sector6.2%

Emerging Market Bonds · rates4.3%

Health Care · sector4.1%

Switzerland · geography2.7%

Crypto names (MSTR, COIN)— none

Holding by holding

weight · vol · top factor · R²

Holding	Reference label	Wt	Vol	Top factor (β, %)	R²
MSTR	US · Info Technology	6.3%	22.8%	none	0.00
COIN	US · Financials	6.7%	22.6%	none	0.00
BABA	US · Cons. Discretionary	6.6%	9.3%	Cons. Discr. 94%	0.23
ASML	Netherlands · Info Tech	6.3%	9.3%	Netherlands 122% + Tech 30%	0.73
TSM	US · Info Technology	6.5%	7.6%	Technology 94%	0.53
AEM	Canada · Materials	6.7%	6.7%	Materials 101%	0.35
NEM	US · Materials	6.7%	6.6%	Materials 79%	0.30
NVO	Denmark · Health Care	7.5%	6.0%	Denmark 124%	0.50
Austria 2117	Austria · Govt · AA+	10.3%	2.8%	Dev Govt Bonds 201%	0.65
UL	UK · Consumer Staples	7.0%	2.3%	Cons. Staples 91%	0.43
NVS	Switzerland · Health Care	6.9%	2.2%	Health Care 59% + Switz 39%	0.57
UST 2050	US · Govt · AA+	10.2%	1.1%	US Treasuries 87%	0.35
PEMEX 2060	Mexico · Corp · sub-IG	7.1%	0.8%	EM Bonds 60%	0.39
USD/NOK	Currency NOK	5.1%	~0%	none	—

Be precise — where it doesn’t diverge

credibility > overreach

Gold miners really are "Materials"

NEM and AEM load on the Materials sector factor; there’s no separate gold-commodity factor here. The label holds — don’t pitch a hidden gold bet.

The bonds really are low-volatility

Reference data calls the sleeve "safe"; on volatility (~5%) it is. The catch is the concentrated duration beta, a scenario risk — not a claim the daily-vol label is wrong.

The European ADRs map to home country

ASML→Netherlands, NVO→Denmark, NVS→Switzerland. Issuer-domicile reference data and the geographic factor agree for these.

Beta is not risk

The headline ranks by contribution to volatility, not factor beta. The largest beta (government-bond duration) is one of the smallest risk contributors. Conflating the two is the easy mistake this exhibit is built to avoid.

How these numbers were produced

reproducible

Built in MantaRisk from reference-vs-factor-demo-portfolio.csv (14 instruments). Factor betas and volatility contributions are MantaCore’s live output (compute_instrument_exposures, LASSO, daily returns since 2010); the reference-data classification is read from the same loaded instruments and weighted by MantaRisk’s own allocations. R² = 0.58, diversification ratio 1.80.

Betas are normalized to a fully-invested basis. Volatility-contribution shares are scale-invariant and taken directly from the engine.