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

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.

That is expected. Each run fetches the latest prices from yfinance and recalculates. When new trading days arrive or the factor data is updated between runs, the sample changes and the means, variances and regression coefficients shift slightly. The risk-free rate used for the Sharpe ratio (the US 13-week T-bill yield) is also refreshed daily. If the change is large (e.g. alpha flips sign, or a beta moves more than 0.3), the sample may be too short for stable results — check the analysis reliability score at the top of the result page.

Holding amounts are converted to KRW at the FX rate at analysis time to set the weights. Returns, however, are calculated from each holding's local-currency price, so the returns and volatility of US holdings do not include KRW/USD exchange-rate moves. If you invest in won, your actual account can swing more or less than shown, by the amount the exchange rate moves. Markowitz and HRP analyze holdings from both markets as one portfolio. Fama–French (3- / 5-factor) regresses Korean holdings on DIVA's own Korea factors and US holdings on factors from the Kenneth French Data Library, then combines the two by portfolio weight into a single result.

A 0–100 measure of how much the numbers could shift because of limited data or chance (not whether the portfolio is good). For Markowitz and HRP it averages ① the risk estimate (data window, and window length relative to the number of holdings) and ② the return estimate (the 95% error range of the past average return), so the same portfolio over the same window scores the same in both. Fama–French averages the beta estimate (window and beta error range) and explanatory power (R²). 75+ reliable, 55+ worth referencing, 35+ read conservatively, below that interpret with care. A heuristic for reference, not an academic standard.

Markowitz optimal weights come from inverting the covariance matrix, so with many holdings or a short window, a small change in the input can produce extremes like 99% in one name and 0% in the rest. DIVA therefore uses the Markowitz result to diagnose how far your current weights sit from the efficient frontier, and leaves weight suggestions to HRP, which needs no inversion and gives stable results. The HRP weights are not the answer either: they ignore expected return and put more weight where volatility is lower, so applying them can lower volatility while also lowering expected return and Sharpe. Run the same portfolio through both models: use Markowitz to see where you stand, and HRP to see what switching weights would gain and cost.

In three steps. ① Holdings whose correlation patterns are similar are linked into the correlation structure tree, based on monthly returns. ② Holdings are lined up in the tree's leaf order. ③ The line is cut in the middle, the two halves' volatility is compared, and more money goes to the less volatile half. Each half is split the same way until single holdings remain. So the tree sets the order of holdings, and the actual cut points can differ from its branches. The allocation at each step is shown in the "How investment weights are determined" tab.

Yes. At every step HRP puts more money into the less volatile half, and repeats the same rule inside each half. So if the portfolio includes calm assets such as bond ETFs or low-volatility dividend stocks, their weights rise sharply while volatile names such as growth stocks shrink. Expected return is not an input to the weights, so a lower expected return under HRP is a natural result. Use the "Metric comparison" tab to judge whether the lower volatility is worth the lower expected return.

Circle area is each holding's share of total portfolio risk (variance) — its risk contribution. Use the buttons above the tree to switch between the current portfolio and the HRP balanced portfolio and see where risk concentrates. Dashed circles and negative values mark holdings that move against the others and reduce total risk. With such offsets the remaining contributions can add up to more than 100%, so the "Largest risk contribution" metric can exceed 100%. Holdings joined at a lower distance move together more closely.

Neither. HRP is not a volatility-minimizing calculation but a rule that follows the correlation structure and puts more weight on the less volatile side, so volatility can even come out higher than under your current weights. Nor does it equalize each holding's risk contribution (risk parity), so contributions still differ after switching. Check the "Under HRP" values in the key metrics together with the HRP balanced view of the correlation structure tree. "HRP reallocation" is the share of capital you would need to move to get there, excluding trading costs and taxes.

It runs, but HRP's advantage barely shows. With two holdings there is a single cut, so the weights are simply inversely proportional to each holding's variance (more in the calmer one), and their correlation has no effect on the weights. The benefit of using the correlation structure to spread risk within and across groups usually appears from 4–5 holdings up.