DIVAQuantizer
HomeModelsAnalyzeDashboardTheoryBlog
← Back to input·Analysis Result
PROHierarchical Risk Parity

예시 포트폴리오 — 미국 분산 (AAPL · MSFT · SPY · TLT)

● 4 holdings· $ 10,000· 72 mo window· 2026.07.08 · 18:00
This is a "sample analysis". It is not based on your own portfolio — we are showing a pre-prepared analysis of 예시 포트폴리오 — 미국 분산 (AAPL · MSFT · SPY · TLT) for learning purposes. To analyze your own holdings, click the button on the right to go to the input page.All numbers were computed against real market data at the analysis time. The snapshot is fixed, so values may differ from the most recent market conditions.
●HRP · Portfolio personality

The Obsessive Researcher집념의 연구가

A single-track mind that digs deep into one field — a narrow scope, but with relentless tenacity you bore straight toward a clear goal.

Key metrics · annualized
Current σ
16.26%
Prices swing an average of 16.3% up and down per year.
Current μ
+14.67%
Past average implies +14.7% return per year.
Current Sharpe
0.68
Earns 0.68 per unit of risk — moderate.
Diversification
19.3%
Diversification has cut risk by 19.3%.
Visualization

Correlation-based asset clusters

0.000.140.290.430.58DISTANCE ↑d = 0.38d = 0.46d = 0.58TLTTLT44%AAPLAAPL12%MSFTMSFT14%SPYSPY30%RISK ANCHOR · 70%CONVERGENT BET · 30%
Read

Holdings that merge close together move together strongly. Lower y-axis = closer distance; far-apart branches act as separate groups.

HRP uses this tree to spread risk within each group while keeping groups themselves diversified. Colored regions above mark the merge range of each cluster.

Interpretation4 takeaways
01 · Diversification
Diversification 19.3% — moderate

Diversification of 19.3% is the normal range for a portfolio that mixes asset classes — stocks with bonds, or domestic with overseas. There's enough low-correlation material in the mix that one leg can cushion another when conditions turn.

02 · Concentration
Rebalance suggested

A max gap of 24.1 pp means one or two positions are doing a lot more lifting than HRP would prefer. When those names struggle, the whole book feels it. Easing TLT (and similar offenders) toward their suggested weights, gradually, will noticeably improve risk balance.

03 · Cluster
2 clusters — clear binary structure

Two clusters, each cohesive — a clear two-layer structure (think growth/tech on one side, defensive/dividend on the other). That's meaningful diversification already. The natural next step, if you want one, is bringing in a third flavor — bonds, gold, or overseas exposure — to add a layer.

04 · Reliability
Near-duplicate assets (1 pair)

Correlations above ±0.95 show up in 1 pair — effectively the same asset held under two names. For diversification, those pairs should be treated as a single position. Clustering will always merge them, which limits how informative the cluster result can be. Removing one of each pair or swapping it for a different-flavored asset typically sharpens the structure.

Rebalance Proposal

Top 3 weight changes

01
TLT
Increase
20.0% → 44.1%
+24.1pp
02
AAPL
Decrease
25.0% → 12.4%
-12.6pp
03
MSFT
Decrease
20.0% → 13.8%
-6.2pp
Next step

팩터 노출까지 함께 보고 싶다면?

Fama–French 회귀로 시장·규모·가치 노출을 분해해 분산 효과의 출처를 검증할 수 있습니다.

FAQ

Frequently asked questions

Questions newcomers to this model commonly ask. Click any question to expand the answer.

Can I trade exactly the way the analysis suggests?
No, that is not recommended. Every analysis here is a statistical estimate based on historical market data, and it does not guarantee future returns or losses. Trading costs, taxes, FX costs, market impact, and your own investment goals and horizon are also not reflected. Use the results as a diagnostic of "what kind of risk and return profile this portfolio has", and before any real investment decision review your own situation and consider consulting a qualified professional.
I re-ran the analysis on the same tickers but the result is slightly different — why?
That is expected. Each analysis fetches the latest data from yfinance and the Kenneth French library at the moment you run it. When new trading days or factor updates land between runs, the sample changes and the means, variances and regression coefficients shift slightly. If the change is large (e.g. alpha flips sign, or a beta moves more than 0.3), that is a signal that the sample is too short for statistical stability — check the reliability score at the top of the result page.
How are Korean and US stocks handled together?
It depends on the model. Markowitz and HRP convert all amounts to a single currency (KRW) and treat the two markets as one portfolio — USD holdings are converted at the FX rate at analysis time. Fama–French (3- / 5-factor) needs region-specific factors, so Korean tickers are regressed against our own Korea factors (computed in-house from public Korean market data) and US tickers against Kenneth French's North America factors, with the two regressions combined by weight average. Mixed KR / US portfolios appear as a "split analysis" on the result page, with per-market regression results included.
How is the "reliability score" calculated?
A 0–100 score combining the sample window (months) with model-specific quality signals. The window component reaches 70+ once it crosses 60 months (5 years, the academic recommendation). Per model: Markowitz adds diversification benefit and a single-name concentration penalty; Fama–French adds R² and the share of betas that are statistically significant; HRP adds diversification benefit and how much HRP reduces current volatility. 75+ is "excellent", 55–75 "good", 35–55 "fair", below 35 "low". This is a heuristic for "how seriously should I take this result" — not an academic standard.
When HRP and Markowitz recommend different weights, which should I follow?
Usually HRP is more robust to small perturbations in the input. Markowitz inverts the covariance matrix, so with many holdings or a short window it can swing widely; HRP is cluster-based and tends to produce more stable weights for the same input. With many holdings on a short window, HRP is the higher-confidence choice. If your priority is efficiency and you trust the inputs, weights closer to the Markowitz recommendation are also worth considering. The best move is to run both and inspect which names the two disagree on most.
Why do clustering results change between runs?
Any change in the window or the holdings shifts the pairwise correlations, which shifts the distance matrix and the cluster tree. Even with the same 5 tickers and 6 years of data, a month later new trading days enter the sample and the clusters can move slightly. To gauge stability, check the dendrogram's vertical axis (distance) — clusters that split at large distances are stable, while names that move between groups when distances are close are sample-dependent.
How are cluster labels like "Group 1 / Group 2" assigned?
The labels just follow the dendrogram's ordering — 1, 2, 3 in order. "Group 1" is not "the most important group". The real information is who is in each group and the average correlation within it. The result page shows, per group: the tickers in it, the average within-group correlation, and the aggregate weight that group represents. Use those to read "how many meaningful risk clusters my portfolio breaks into".
Does HRP always produce a larger diversification effect than Markowitz?
Usually similar, often slightly larger for HRP — but not always. HRP allocates risk evenly across and within clusters, which naturally yields strong diversification numbers. Markowitz, under the same mean-variance assumptions, can in theory land on a point of the efficient frontier — if you trust the inputs, it may yield a more efficient point. The objectives differ: HRP optimizes for "robust diversification", Markowitz for "efficiency". Looking at both side by side lets you weigh each strength.
Is HRP meaningful with only 2 holdings?
Technically yes, but the clustering insight is thin. With only two names the tree is a single step, so HRP's core "within vs across cluster" framing does not really show up. The tool allows 2 holdings as a minimum, but the robustness advantage typically becomes visible from 4–5 holdings onward. With few holdings, HRP results often coincide with Markowitz, and the comparison between the two models loses value. More holdings give a richer cluster tree and more meaningful HRP weight differentiation.