Cypress Investment Research: how AI reads diversification and correlation
A free reading desk on one subject: how AI estimates what moves together, and what that does to portfolio diversification. Articles only — nothing is sold, and the decisions stay yours.
Questions cost nothing either — call +886 2 2716 4823 or write info@cypress-module.digital
What every article is asked to contain
Diversification lives or dies on correlation — on how pairs of holdings behave when one of them shakes. Machine learning changed how that question is estimated, not whether it deserves a hard look. Our articles give it one.
- The metric under the glass is named early — a rolling correlation window, a clustered covariance estimate, a factor exposure breakdown — in the first screen, never buried in a footnote.
- Every chart reruns from public market data the reader could fetch, with prices, window and index stated plainly in the text.
- Where the estimate wobbles — short histories, stale prices, a regime that just broke — is written in as a limit, not airbrushed as smoothness.
- Claims about what “AI” does are pinned to the method that did the work, cited inline to its published source.
- No article names a position to buy or sell, and none is written for one portfolio — the scope line at the top of each piece says so twice if needed.
Three threads the desk keeps pulling
Short, unglamorous, and renewed slowly — the point is that each thread can be checked line by line.
Portfolio diversification, rebuilt by clustering
Classic diversification counts tickers; machine-learning methods count behaviours. Articles here re-run clustering passes over index universes and show when ten holdings quietly behave like four.
- hierarchical grouping of assets by return behaviour
- concentration metrics that catch hidden overlap
- tail spreads when correlations converge under stress
Correlation analysis, window by window
Correlation is not a number; it is a windowed estimate with opinions. We chart how the estimate moves and when it stops meaning anything.
- rolling and exponentially weighted estimators, side by side
- regime shifts that quietly re-price diversification
- spillover between markets that trade while Taipei sleeps
Assumptions under the code
Every estimator carries assumptions in its pockets. We unpack shrinkage and factor models against raw sample covariance — and ask what breaks first.
- shrinkage vs. raw covariance, in plain arithmetic
- walk-forward checks a backtest actually passes
- when a neural network is the wrong tool for the job
The Taipei exchange closes at 13:30; the correlations it leaves behind run all day
Written a short walk from the TWSE quotation wall, our articles trace how one session in Taipei ripples into overnight books elsewhere — spillover, made local.
Man Ng · Wikimedia Commons · CC BY 2.0
How a piece gets written, in five moves
No model runs at this desk — we write about them. The writing itself stays deliberately manual.
- Frame one narrow question — say, whether clustering a bond-plus-equity universe really reduces hidden overlap, or just repaints it.
- Pull public data a reader could fetch themselves — daily closes, index weights, exchange disclosures — and state the window up front.
- Redraw the indicator, by hand where possible: the rolling correlation matrix, the dendrogram, the factor loadings.
- Write the limits paragraph before the conclusion — what window choice, data gap or regime shift could flip the answer.
- Name every source in the text, and label whatever remains speculative as exactly that, reasoning attached.
The desk is a publisher, not a fund: no capital moves through this site, no accounts exist here, and nothing on it is for sale.
Who this reading is for — and who it isn’t
No invented praise, no borrowed ratings: just an honest fit-check so nobody wastes a bookmark.
…want the method named, not the adjective
- you compare claims about AI in portfolio diversification and want to see the actual estimator
- you manage your own money and want plain-English models of correlation risk to think with
- you study finance or data science and want worked examples between the papers and the press releases
- you suspect a correlation heatmap can mislead as easily as it informs, and want to know exactly how
…need a signal
- you want buy or sell calls — nothing here will name a position
- you need personalised investment advice — we cannot give it, in any phrasing
- you are shopping for a fund, signals or a subscription — none exists here
…always in force
- no paid services, products or memberships, at any price
- no execution, no custody, no deposits — no exceptions
- inquiries about methods and sources are the only thing on the counter
Ask the desk to unpack any diversification claim
The inbox is the whole service counter — no funnels, no retainers, no follow-up sequence. Request a topic, challenge a chart, or ask what sits under a correlation number that looks too tidy.
Ken Marshall · Wikimedia Commons · CC BY 4.0