In the early weeks of a bond issuance or a project financing, a recurring scene plays out before any agency has been mandated: the finance team holds separate preliminary conversations with three or four rating agencies, presenting the same financial model, the same contract package and the same sponsor structure to each, and comes away from every conversation with a verbal indication of range. Three of those ranges cluster closely together while one sits a notch above them, and the mandate, in all likelihood, goes to the agency that produced the outlier. Nothing improper has occurred at any step; the information provided was identical, the conversations were conducted separately and transparently, and the selection was presented to committee on grounds of methodology fit and sector experience. Yet the probability that the final published rating equals the highest preliminary indication is structurally higher than the probability that it equals the median of those indications.
Viewed from the investor side, the same pattern presents itself as a distributional artifact rather than as an individual decision. When assets carrying two agency ratings are placed alongside assets carrying one within a single credit portfolio, the single-rated group tends to concentrate toward the upper end of the distribution, and the absence of a second rating is not a gap in the file but the residue of a choice. Because the indication offered by the agency that was ultimately not mandated is never reduced to writing and therefore never enters any file, a treasury team sitting down to a refinancing two years later has no awareness that such an indication ever existed. What survives in institutional memory is the name of the selected agency and a single line of rationale referencing methodology.
The behavior has a name — rating shopping, the practice of mandating the agency expected, on an identical information set, to deliver the highest rating — and its mechanism rests on two supports. The first is the issuer-pays model, under which the cost of the rating service is borne by the party whose economics are most directly affected by the outcome. The second, and the more consequential, is that methodological divergence among agencies is real and legitimate: agencies differ on whether cash flow is treated as availability-based or volume-based, on how parent support is notched, on the weight assigned to construction-period risk, and on the extent to which residual value is recognized, and that divergence produces several defensible ratings for the same asset. Rating shopping does not create the distribution; it selects its upper end and renders the remaining points invisible.
The mechanism functions legitimately under identifiable conditions, and recognizing those conditions matters, because an intervention aimed at the wrong target does damage of its own. An agency with genuine depth in a particular asset class may well capture that asset's economics more accurately through its criteria set; a rating from an agency with established standing in a sector may reach a broader investor base in secondary trading; process cost and calendar are, on their own, legitimate selection criteria. The difficulty lies not in the shortcut itself but in the silent substitution of one criterion for another. When the selection begins with the question of which methodology best describes the asset and drifts toward the question of which agency will produce the higher rating, the same decision continues to be defended on the same grounds while optimizing for something different. The drift appears nowhere in writing, because the text of the rationale never changed.
That drift produces a second effect inside the issuer's own internal model. The highest indication heard during the preliminary phase becomes an anchor for the team's own expectation, all subsequent internal discussion is constructed around that notch, and more conservative indications are reinterpreted as exceptions — sometimes as evidence that the agency in question misunderstood the asset. The width of the distribution, which is the quantity actually carrying credit information, is thereby dropped entirely from the output of the process. What remains is a single number representing the maximum of a distribution, whereas what matters in a credit decision is the distribution itself.
The first cost of this configuration surfaces in pricing, and it is generally smaller than anticipated — which is precisely the point. Institutional buyers, having observed single-rated issues over extended periods, typically read a rating formed at the upper end of a plausible range with an implicit notch discount and calibrate their required yield accordingly. The issuer therefore pays both a direct fee and process time for the incremental notch while recovering only part of it in spread. Measured against transaction economics alone, this is a marginal loss and not, on its own, large enough to distort the structure. The material cost accumulates elsewhere.
That cost sits in the trigger inventory embedded in the capital structure. Rating conditions in credit agreements, collateral posting obligations under derivative documentation, credit support provisions in long-term offtake contracts, terms governing letters of credit and performance bonds, index eligibility rules, and the internal limit definitions maintained by institutional investors — a significant share of these cluster immediately around the investment-grade boundary. The cushion carried by a rating two notches from that boundary is not the same instrument as the cushion carried by a rating one notch away; in the latter case a single downgrade can activate several obligations at once, and the resulting cash demand is stepped rather than linear. The notch obtained from the upper end of the distribution was purchased, in substance, by consuming exactly that cushion.
The second cost becomes visible at the moment the asset changes hands or is refinanced. A buy-side diligence team applying the same data set to the published criteria of the agencies that were not mandated obtains a distribution, and the gap between the standing rating and the median of that distribution converts into a line item in the acquisition negotiation — typically expressed as an escrow ratio, a condition precedent, or an undertaking tied to rating maintenance. At that point, to the extent the issuer holds no record demonstrating that the original selection rested on methodology fit, defending the gap becomes considerably harder. The file contains the outcome; it does not contain the reasoning.
The mechanism that neutralizes this tendency is built into process architecture rather than individual attentiveness, and it separates into four components. The first is recording the mandate decision at the moment of proposal rather than at the moment of approval: which agencies were nominated and against which criteria, which indications were received from preliminary conversations, and which specific methodology provisions the selection rested on, all committed to writing before the decision is taken. The second is criteria mapping conducted in advance of any preliminary conversation — establishing which agency's criteria set corresponds to the asset's economics before a single indication has been heard, which removes the ground on which the anchoring effect would otherwise form. The third is a shadow rating run against the published criteria of every candidate agency, producing a distribution rather than a single number. The fourth is a threshold policy that defines a minimum notch cushion against every rating trigger in the capital structure and makes that cushion a binding constraint on the mandate decision itself.
In operating this architecture across capital-intensive and financed projects, BEIREK maintains three records and runs one recurring discipline. The first record is the trigger inventory: credit agreements, derivative documentation, long-term offtake contracts and security structures are reviewed so that every rating-linked obligation is consolidated into a single schedule, with each trigger's distance from the current rating measured in notches. The second is the mandate file, in which every indication received during preliminary conversations — including those given verbally — is recorded with its date and its stated scope, so that the width of the distribution survives into the output of the process. The third is a reasoning memorandum identifying where the selected methodology converges with, and where it departs from, the actual cash flow mechanics of the asset.
The recurring discipline is the construction of the downgrade path in reverse before closing: which scenario breaks which criteria provision, which triggers that breach activates and in what sequence, and from which source the resulting cash demand is met, written once before the transaction is signed and refreshed at intervals tied to the rating review calendar. Within that discipline sits a distinct role charged with arguing the opposite case, whose function is to defend the criteria set of the agency that was not mandated and to test whether the selection remains defensible against those criteria as well. The quality of a rating is measured not by where in the distribution it was drawn from, but by whether the width of that distribution was recorded at all; an unrecorded distribution presents itself, unrehearsed, at the first rating review.