Spot holdings that are really the same bet and adjust their sizes to allow for it.
You can give every holding a single-stock limit, every sector a limit and every country a limit, check each one, and still own one large bet that nothing has sized. This happens when several holdings depend on the same underlying force: interest rates, oil, the chip cycle, regional growth or the US dollar. The holdings can sit in different sectors, trade on different exchanges and have different names, so the per-holding limits never catch them. A group of holdings that react to the same underlying force, so that in a shock they behave like one position, is a correlation cluster.
Lesson 2.4, Beta, correlation and why diversification shrinks in a crisis, showed that pairs with a correlation above 0.8 behave like one position. The next step is to turn that observation into a sizing rule.
Marcus, the course's made-up investor, found this in lesson 4.8, Run a geopolitical scenario through your holdings. When he ran an oil shock through his portfolio, his bank, his STI ETF and his S-REIT moved together, and he noted that he would set a combined limit for the three in his risk rulebook. Lesson 11.3, Position and sector limits, and concentration risk, had already given him single-stock, sector and country limits. The three mostly passed those limits one at a time, but together they were one unsized bet.
Start with a list of your holdings and their values. For each one, ask what one thing going badly would hurt it most, and write that answer beside it. Holdings with the same answer form a cluster. Group by driver, not by label such as listing exchange, country or sector category.
Marcus's answers gave him four clusters. His holdings, weights and most of the correlations used here are made-up examples; only the two correlations taken from lesson 2.4 are real.
Global shares: the world ETF, about 45% of the portfolio. Singapore interest rates: the SGD bond fund and T-bills, about 19.8% together. Singapore rates and regional growth: the STI ETF, the bank and the S-REIT, S$48,000, about 23.8%. The chip cycle: the chip designer and Larkspur, S$23,000, about 11.4%.
The third group shows how different businesses can share one driver. The bank is driven by Singapore rates and regional growth through its lending margin and loan losses. The S-REIT is driven by the same forces through its borrowing costs and property values. The STI ETF follows them too, because banks are its largest weight.
Larkspur shows why the label misleads. It is listed on SGX and categorised as Singapore precision engineering, so a label-based view would put it with the Singapore holdings. Its main driver is chipmakers' spending. In a chip downturn it would fall with the US chip designer, and Singapore banks would not take it down with them. Marcus placed it in the chip cluster.
Correlations from your own price data help confirm the grouping. The bank and the STI ETF have a correlation of about 0.88, and the chip designer and the world ETF about 0.82, both from lesson 2.4 and both above the 0.8 line. Marcus's made-up figures back his choice for Larkspur: its monthly returns have a correlation of about 0.7 with the chip designer and about 0.4 with the STI ETF. Lesson 4.3, Semiconductor supply chains and export controls, also found Larkspur and the chip designer moving together on the day new export rules were announced.
Where your data and your judgement disagree, ask why before trusting either.
A cluster limit is a maximum share of the portfolio for any group sharing a driver, counted as a single position. Size each cluster as if it were one holding. Marcus set his at 20%. Core world equity and bond funds are exempt, because their sizes are set by the asset allocation, so his world ETF and his bond fund and T-bills are left out.
Next, total each cluster and compare it with the limit. Marcus's Singapore cluster, at 23.8%, breaks the 20% limit. The fix does not have to come only from selling the obvious name. Trimming the bank to its 5% target from lesson 11.3 brings the cluster to about S$42,100, or 20.8%, which is still over. Trimming the S-REIT to its 5% target as well brings it to about S$40,200, about 19.9%, just inside. Both trims also help the Singapore country limit from lesson 11.3, so check knock-on effects like this whenever you trim.
His chip cluster, at 11.4% before any trims, passes. Its look-through total, which counts the chip exposure held inside the world ETF as well, sits on the sector limit, so trimming the chip designer to its 4% target helps there.
Correlations are measured in ordinary months, and ordinary months flatter them. In a market-wide fall, investors sell whatever they can, so holdings that moved separately in calm years fall together. Lesson 2.4 showed this for 2008 and March 2020. Treat calm-period correlations as a minimum. Marcus's S-REIT and STI ETF have a correlation of about 0.6 in calm years, and as an illustration it could become 0.9 in a crisis. If correlations rise toward 0.9, a cluster's combined loss is close to the sum of its parts rather than smaller.
The crisis loss of a cluster is what the cluster limit is really about. In lesson 11.6, Stress test against past crises and made-up shocks, Marcus's made-up 2020-style fall had his STI ETF, bank and S-REIT together lose about S$15,500, about 7.7% of the whole portfolio. Run your own clusters through a crisis scenario the same way.
One cluster cannot be sized away: shares as a whole. In a severe crash almost every equity holding joins it. The cluster method has not failed here. This is the equity risk you chose through your allocation, which for Marcus is 80% in equities, and it is handled by your allocation and rebalancing rules, covered in the course Build and run an ETF portfolio, lesson 6.2, Calendar or threshold rebalancing.
Finally, write the cluster rule into your rulebook. Marcus wrote: "Any group of holdings sharing one main driver is capped at 20% of the portfolio, counted as if it were a single position; core world equity and bonds are set by my allocation instead."
Open your list of holdings with their values and get ready to write beside each one the single force that would hurt it most.
Sort your holdings into clusters by main driver, add up each cluster's weight and compare it with your position limit.
Junxiong-WFG Organisation is an authorised representative of AIA Financial Advisers Private Limited (Reg. No. 201715016G).