Every portfolio-level DV01 or CS01 is the sum of position-level numbers, and every position-level number is the output of a calculation that made assumptions for curves, spread definition, pricing, modified or effective duration, option adjusted spreads or simple government curve. The portfolio number is the sum of all those choices.
In a well-governed shop, the methodology is documented and applied consistently. In most shops, the methodology is consistent within an asset class and inconsistent across them. The credit team uses one curve set, the rates team uses another, the structured book uses something else, and the multi-strategy fund inherits the average of all three when the risk report runs after the close.
This becomes obvious in stress testing. Move spreads 50bp wider and the portfolio shows a P&L impact of X. Run the same scenario from a different starting curve and the answer changes by 8%. The risk number was always sitting on a methodology choice and stress testing surfaces those choices.
There is a related problem with concentration. CS01 by issuer is one of the more useful credit risk views because it shows the source of spread sensitivity. But CS01 by issuer requires a clean issuer mapping, and clean issuer mappings are inheritely complex datasets. Subsidiaries roll up to parents in some hierarchies and not others. For example:
The takeaway for a credit PM is not to distrust the risk system. The system is just doing what it is told; it is the configuration that needs to be understood. When the chief risk officer asks why portfolio CS01 looks lower than last quarter, the right answer is rarely "the market moved". It is much more likely that the "the curve assumptions changed in February" or "the issuer mapping was updated when the new ratings file was loaded" and those answers do not show up on final slides for the risk committee. The audit log is rarely an appendix
Two PMs running notionally identical portfolios will publish different DV01s if their methodologies differ. Two systems analysing the same portfolio with the same data will publish different CS01s if their assumptions differ. This is frustrating because it is solvable and avoidable. A consistent curve hierarchy, a maintained issuer mapping and parent rollup, a single source of pricing assumptions, and a documented stress methodology produce comparable numbers.
The reasons these discrepancies keep creeping into power bi or tableau dashboards of the quant analysts who puts the final deck together is not a lack of technical understanding, it is a lack of model governance and model ownership which results in a general distrust in the risk analytics.