The diversification trap: more holdings ≠ more diversified
One of the most common investor illusions: "I own many holdings, so I'm diversified." Maybe — but often not. Semiconductors, platforms, EVs, growth names — when the market starts to fear rising rates, they can all fall together. The names look different, but the underlying risk was the same.
HRP is built to catch this illusion. Instead of trusting tickers or sector labels, it looks at how prices actually co-moved, and groups holdings whose movements resemble each other. Treat the portfolio as a map of risks, not a list of names.
The goal isn't to predict "the next big winner." It's more grounded: check whether your portfolio is betting too much on a single story, and spread risk more evenly. Survival before prediction. Think of it as a helmet — not stylish, but it keeps your head intact.
"HRP replaces the covariance structure with a tree structure. In doing so, it prevents small estimation errors from leading to entirely different solutions."
— Marcos López de Prado, Building Diversified Portfolios that Outperform Out-of-Sample (2016)
The 3 things to read first
Read the structure before the numbers. These three together give you the layout of your risks at a glance.
01 · Cluster
Which holdings are on the same team
Holdings that co-move land in the same cluster. A crowded cluster means you carry a lot of that particular risk.
02 · Weight
Where does the weight pile up
Look at weight per cluster, not per holding. One dominant cluster = de-facto concentration, regardless of holding count.
03 · Surprise
Any surprising groupings
Unexpected pairings are the signal. The market may be treating those holdings as a similar risk, even if your intuition says otherwise.
The best part of HRP is the surprises. A name you thought of as a consumer stock landing in the tech cluster means it actually moved like a growth play. You won't see these from a flat weights table.
HRP organizes risk in three steps
It looks like dense math but the idea is simple: measure distance → group similar names → split risk between groups. That's it.
Example · 5-holding cluster tree
① correlation → distance
AAPLHDKOZGSams.AAPLHDKOZGSams.
Darker = moves together = closer
② cluster similar names
③ allocate per cluster
Final weights
AAPL22%
HD18%
KO25%
ZG15%
Sams.20%
01
Distance · measure proximity
The more they co-move, the closer they sit
High correlation = nearby in HRP's view. If one rises while the other is unmoved (or moves differently), they sit farther apart.
How to
Movements matter more than names. Markets speak with prices, not labels.
02
Tree · group into clusters
Nearby holdings get grouped into clusters
Holdings that move alike merge onto the same branch. The tree maps the hidden structure of your portfolio.
How to
Fewer clusters often means fewer distinct risks. Many names + few clusters = diversification illusion.
03
Allocation · split risk recursively
Split between major groups first, then within each
HRP doesn't optimize all holdings at once. It splits risk between major clusters first, then within each cluster — making it harder for any one name to dominate the weights.
How to
Stable allocation over extreme bets. Less thrill, more sleep.
5 ways to actually use HRP
Don't stop at reading the result — HRP works best as a checklist for fixing the portfolio.
01
Reality Check · spot diversification illusion
Check whether you're actually spread across different risks
Many holdings but only 2~3 clusters? Real diversification is weak. The fix is not to buy more — it's to find assets that move unlike your existing clusters.
How to
Count clusters → check the largest cluster's weight → screen for candidates moving differently from it.
02
New Asset · validate candidates
Is the new name a new team, or just another seat at the same table?
If your candidate joins an already-large cluster, the diversification gain is limited. If it opens a new branch or reinforces a small cluster, structure improves.
How to
Re-run with the candidate → see which cluster it joins → decide: does it bring new risk, or just amplify existing risk?
03
Rebalancing · structural, not name-by-name
Trim the team that grew too big first
Just selling the winners is blunt. HRP-style rebalancing happens at the cluster level: if one team has ballooned, trim within it and shore up the under-weighted teams.
How to
Compare current vs HRP-suggested weights per cluster → adjust the biggest gaps first → decide per-name trades last.
04
Model Risk · sanity-check other models
Use HRP as a brake when your equalizer weights run hot
Push the Markowitz equalizer toward a higher Sharpe and weight tends to pile onto the one or two names with the best past returns, which is often the expected-return estimate doing the work. HRP weights ignore expected return, so they serve as a structural check on whether that tilt makes sense.
How to
Note the names you raised in the equalizer → compare with the HRP weights → if yours are far above HRP, cap them or scale in over time.
05
Stress Mode · noisy markets
Lean on HRP as the default when the market is noisy
When rates, FX, recession risk, geopolitics all cloud the read, predictions weaken. HRP gives you a baseline that aims to "break less," not to "win more."
How to
Volatility spikes or unclear outlook → set HRP weights as the baseline → only over-weight names you have strong conviction in.
A good helmet doesn't steer the car
HRP is a safety device. It doesn't replace direction, profit forecasts, or market judgment.
!
Don't treat HRP as an "answer generator."The better use is to surface the concentrations in your current portfolio and to sanity-check whether other models are running too hot.
HRP doesn't lean on expected returns. So even if you're confident a name will rip higher, the model won't reflect that view automatically.
Limitation 02
Correlations can break down in a crisis
Assets that move differently in normal times often sell off together in a crisis. Don't assume past clusters survive the next regime.
Limitation 03
Cluster interpretation still needs human judgment
The model groups but doesn't explain why. Is it rates? The dollar? A shared theme? That part is on you.
Limitation 04
Can feel too conservative
HRP dislikes extreme weights. If you want high-conviction concentration, the result can look bland. That's both its strength and its weakness.
You've seen the risk map — now look at the source of returns
HRP shows you the structure of the portfolio. But where the returns came from is a different question — broad market beta? exposure to size or value factors? actual selection skill? — and needs a separate look.
Recommended Next
Fama-French 3-Factor analysis
Decompose returns into market, size, and value premia. HRP showed how risk is arranged — 3-Factor shows where the returns are coming from.