A large customer count does not establish diversification. Hundreds of separate companies may depend on the same funding market, major buyer or collateral price. My starting question for concentration analysis is therefore about a shared cause: what could go wrong for several customers at once, and how would that movement reach the bank? This article combines a documented overseas case with my proposed analytical approach. It does not describe an inspection I conducted or assess a particular Czech bank. The purpose is to turn an abstract diversification question into a sequence of steps that another analyst could examine and reproduce.
Documented case: Silicon Valley Bank in California
Silicon Valley Bank was closed in California on 10 March 2023. In its review published on 28 April 2023, the Federal Reserve described a highly concentrated business model, reliance on uninsured deposits and weaknesses in risk management. The customer base was closely connected to technology companies and venture capital. The review also examined interest rate and liquidity risks and shortcomings in supervision. It is therefore misleading to reduce the case to a claim that exposure to one sector always causes failure. It illustrates interacting dependencies and other weaknesses. The primary sources below provide the documented chronology and findings.
My proposed connection to credit portfolios
SVB is principally a case about the business model and funding, rather than a simple example of corporate loan concentration losses. The connection to lending is my analytical inference: borrowers, like depositors, may share dependencies even when they are separate legal entities. I would distinguish legal connections, economic dependence and common sensitivity to an external shock. These perspectives can produce different groups. A common sensitivity does not automatically imply identical regulatory treatment. The analytical map should first explain the transmission mechanism and then support decisions about limits or further investigation. That distinction prevents a historical case from becoming an unsupported conclusion about another bank.
Define the calculation before measuring it
I would begin with the reference date, exposure perimeter and purpose of the output. Drawn lending produces a different picture from committed facilities including undrawn amounts. The report must state whether it uses gross exposure, a measure after collateral or another quantity. I would avoid mixing those measures within one percentage. Next come identifiers, currencies, duplicates and the mapping between facilities and customers. Without a stable base, a change in the result may reflect a data correction rather than new risk. Every report should therefore include a short statement of scope and known limitations in the input data.
Several views instead of a single definitive table
I would build separate views by sector, geography, exposure size, repayment source and collateral type. Each should show both amounts and percentages. A sector’s share can fall while its loan balance grows if the rest of the portfolio expands faster. Unknown information deserves its own visible category. Hiding it within other items may improve a chart’s appearance while weakening its meaning. The aim is not to create as many tables as possible. Each view should answer a specific question and identify who can verify the underlying information for a material concentration, including when those records were last updated.
Hypothetical example: ten suppliers, one buyer
The following situation is illustrative. Ten companies in different sectors borrow from the same bank. Some manufacture components, others provide logistics and others maintain equipment. Each has separate management and ownership, but all earn a substantial share of revenue from one automotive manufacturer. A sector table may look diversified while a loss of that buyer would affect several borrowers together. My proposed analysis would first verify the buyer’s contribution to their cash flows. Knowing that it is one customer is insufficient. The ability to replace orders, the time needed to transition and available liquidity buffers are more informative than the existence of a commercial relationship alone.
I would separate the initial effect from subsequent amplification. The first effect might be a delayed payment by the manufacturer. Later effects could include production cuts, drawings on reserve facilities and weaker resale values for specialist equipment held as collateral. An analysis focused only on higher default probability could miss the simultaneous pressure on collateral and funding. Yet these channels should not be added mechanically: the same loss must not appear under several names. Before quantification, I would create a short dependency map distinguishing supported relationships from assumptions introduced specifically for an adverse scenario. This makes disagreement about the result much easier to investigate.
Working with incomplete information
A bank cannot necessarily identify every customer of every borrower. My proposal therefore does not demand an apparently complete network of commercial relationships. For material exposures, I would use information available from origination and periodic reviews, record its age and assign responsibility for missing items. An unknown dependence should not silently become zero. A range or sensitivity to a reasonably adverse assumption can support a decision, provided its meaning is explained. Uncertainty about a relationship is not proof of high risk. It may nevertheless justify targeted information gathering before an exposure is increased, particularly when several plausible channels point toward the same dependency.
Turning a report into a decision
The most useful output explains what management should do differently. Options include collecting data, developing a targeted stress scenario, reviewing a limit or monitoring selected borrowers more closely. A concentration does not invariably demand an immediate restriction on lending. Decisions should also consider returns, customer quality and feasible ways to reduce the risk. A proposed sale of exposures requires evidence about market access and execution time. A plan that assumes calm conditions may fail during the shared shock. For each proposed response, I would state an owner, a trigger and major execution constraints, rather than treating a list of possible actions as proof of preparedness.
Explain changes and defend the result
For the next monthly report, I would request a movement analysis separating new business, repayments, currency effects, reclassification and data repairs. Otherwise it is difficult to tell whether diversification actually improved. Alongside the largest groups, I would monitor smaller segments growing quickly. The analysis owner should explain several concrete items behind each material aggregate. In my view, that walkthrough is stronger than checking the spreadsheet formula alone: it tests classification meaning, relationship completeness and the connection to decisions. The output should be reproducible from a retained data snapshot and documented grouping rules, with material judgement calls visible to the reviewer.
Summary and recommendations
I would analyse concentration as a network of shared causes, alongside the familiar ranking of large names. Start with a clear exposure perimeter, compare several purposeful views and make unknown relationships visible. For material dependencies, describe shock transmission and identify a realistic decision with an owner. SVB highlights the importance of a common business environment; specific lending conclusions still need evidence from the portfolio being assessed. A useful analysis therefore goes beyond a warning story. It allows a reviewer to trace a calculation, challenge an assumption and decide which additional information or management change would deliver the greatest practical benefit.