Turning Borrowing Prices into a Map of Uncertainty
The focus now is the reverse direction of credit pricing: given the value of a loan, what promise supports it, and what can that relationship reveal about collateral uncertainty? The answer matters because a credit surface contains more than a quoted spread. Under the model’s assumptions, its complete shape can be translated into the dispersion of collateral returns and into the probability that a loan of a given value becomes insolvent.
From loan value back to the promise
The loan pricing function takes a promised repayment j and returns the lender’s forward loan value:
L(j) = E[min(j, X)].
It answers a promise-centered question: if the borrower writes a promise of j against uncertain collateral X, what is that promise worth to the lender? But credit markets also pose the question in the opposite direction. A borrower may want to raise a particular loan amount, or an analyst may want to compare loans that supply the same value. What promise is required to produce that loan value?
Because L is increasing over the relevant range, it can be inverted. The inverse function J(ℓ) gives the promise that produces a loan value of ℓ:
L(J(ℓ)) = ℓ.
This is an economic change of viewpoint, not just an algebraic operation. L starts with a promise and calculates the expected value of the collateral-limited payoff. J starts with the value the lender is willing to supply and calculates how large a promise must stand behind it. If collateral is more uncertain, the same loan value may require a different promise because the lender expects a different pattern of recoveries.
To make leverage ratios easy to read, the paper normalizes expected collateral value to one. Under that normalization, the loan value ℓ is the loan-to-value ratio: the amount raised relative to the collateral’s expected forward value. The promise j is the corresponding promise-to-value ratio. The loan credit surface is then
S(ℓ) = J(ℓ)/ℓ − 1.*
At each ℓ, the surface tells us how much larger the required promise is than the loan value supplied. A zero spread means that the promise is fully supported in the relevant states. A positive spread means that the promise exceeds the expected value of the collateral-limited repayment by a proportion represented in the ratio.
The inverse function keeps a crucial distinction visible. The horizontal coordinate is the loan value, or LTV, while the promise function supplies the promised repayment associated with that coordinate. The borrower may think in terms of the amount needed, while the lender prices the shortfall risk embedded in the promise. The surface connects those perspectives without treating them as the same quantity.
A simple numerical comparison shows why the inverse matters. Suppose one collateral distribution gives a promise of 100 an expected loan value of 90. The lender must advance 90 against a promise of 100. If a second distribution gives the same promise an expected loan value of 80, raising a loan value of 90 under that distribution would require a promise larger than 100. The difference is not explained by the face amount alone. It comes from the different collateral outcomes that stand behind the same promise.
That comparison also reveals why a surface can be more informative than one spread. One quoted spread might tell us the price attached to a particular loan. The function J(ℓ) tells us how the required promise changes as the loan amount changes. The surface S*(ℓ) then shows how that promise-to-loan ratio varies across the entire leverage range. The information is distributed across the curve.
Why scale disappears
The credit surface is designed to ignore a common change in units. If every possible collateral value and every promise are multiplied by the same positive constant, the ratios used to calculate the spread do not change. A collateral pool worth 100 in one set of units and 200 in another has the same surface if its outcomes have been scaled proportionally and the promises are scaled with them.
This invariance means that the surface does not describe the absolute level of collateral. It describes how collateral value is spread around its mean. The paper calls the distribution of normalized excess returns the dispersion of collateral:
X/E[X].
The distinction between level and dispersion is easy to miss. Consider two hypothetical collateral distributions. In the first, collateral is worth either 80 or 120 with equal probability, so its mean is 100. In the second, it is worth either 60 or 140 with equal probability, with the same mean but a wider range of outcomes. The second distribution has more dispersion even though the average collateral value has not changed. A promise of 100 is repaid as 80 or 100 in the first case, producing an expected loan value of 90. In the second, it is repaid as 60 or 100, producing an expected loan value of 80. The larger downside shortfall changes the spread and the surrounding surface.
Now scale both examples by two. The first becomes 160 or 240, and the second becomes 120 or 280. Means and promises also double. The dollar amounts change, but the relative pattern of outcomes around the mean is the same, so the credit-surface ratios are unchanged. What the surface responds to is not whether collateral is measured in hundreds or thousands. It responds to the shape of the normalized outcomes that determine the chance and size of shortfall.
This explains why a simple fall in the level of collateral is not, by itself, the uncertainty change studied by the paper. If every future collateral outcome falls in the same proportion, the distribution has been rescaled rather than reshaped. Old loans may become harder to honor because their collateral is worth less, but a new borrower who scales the promise and loan value by the same factor faces the same normalized credit terms. A change in dispersion is different: it changes the relative width and downside structure of possible collateral values.
The normalization also determines what the revelation claim can and cannot mean. The surface can identify the dispersion of collateral returns, but it cannot identify the unnormalized scale of the collateral from ratios alone. Two collateral systems with the same normalized distribution but different expected values can generate the same credit surface. The surface is therefore a map of relative uncertainty, not a balance-sheet valuation of the asset pool.
The return arrow represents the paper’s central information result: the surface and the normalized dispersion are two ways of encoding the same uncertainty under the model’s conditions. The diagram is not claiming that an observed spread automatically reveals every economic risk. It shows the one-to-one mapping established within the collateral-pricing framework.
What the surface reveals
The paper proves that the loan pricing function fully reveals the relevant distribution. Its right derivative is the solvency probability, so knowing how L changes with the promise reveals the cumulative distribution of collateral values. Since J is the inverse of L, the promise function carries the same information. And because the loan credit surface is built from J and ℓ, the surface also carries that information.
This is a stronger statement than saying that a spread is correlated with risk. A correlation says that two quantities tend to move together. The one-to-one result says that, under the stated assumptions, the entire pricing relationship can be used to recover the entire normalized collateral dispersion. A single point cannot do that because many distributions can generate the same spread at one leverage level. The full curve supplies the variation needed to distinguish them.
The result is analogous in spirit to using a complete schedule of prices to learn about the payoffs those prices value. It is not necessary to rely on a metaphor, however. The mechanism is already visible in the derivative. At a low promise, the marginal increase in loan value is tied to the probability of full repayment. As the promise moves through the possible collateral values, that marginal probability changes. The pattern of changes across promises is a fingerprint of the distribution.
The loan surface organizes the same information by loan value rather than by promise. For every loan value ℓ, J(ℓ) tells us the promise that reaches it. The surface compares that promise with ℓ. Reading across the surface therefore answers two linked questions: how much loan value can be raised at a given promise, and how costly it is to raise more loan value against the same normalized collateral.
This is why the surface is a better language for the shape of risk than a single spread. If two collateral distributions produce the same spread at one LTV, that equality does not imply that they are equally uncertain. Their curves may diverge at other loan values. One may produce a modest increase in the required promise as ℓ rises, while the other may require sharply larger promises at high ℓ. Only the relationship across leverage exposes that difference.
There is also a practical interpretive payoff. If the surface shifts or steepens after an uncertainty shock, the change can be read as a change in the priced distribution of collateral outcomes. A higher surface says that more promise is required for the same loan value. A steeper surface says that this extra requirement grows more strongly along the leverage dimension. The surface is not merely reporting a lender’s final price; it is organizing how uncertainty enters the pricing of different borrowing amounts.
The dual distribution: default viewed from the loan side
The original distribution F describes possible collateral values. The dual distribution changes the question. It is defined by evaluating the normalized collateral distribution at the promise associated with each loan value:
F*D(ℓ) = FD(J(ℓ)).
The result is the insolvency probability of the loan that raises ℓ. If the borrower raises a larger loan, the associated promise J(ℓ) is larger. The dual distribution asks how likely the collateral is to fall short of that particular promise. It is therefore indexed by the amount of loan value, not by an arbitrary promise detached from the amount raised.
This representation makes the lender’s marginal calculation easier to interpret. The survival probability of the loan is 1 − F*D(ℓ). It is also the marginal increase in the promise function: the extra promise needed when the loan value rises slightly is governed by the chance that the original promise remains solvent. In that sense, the dual distribution translates the geometry of J into a sequence of default probabilities along the LTV axis.
The dual view becomes especially useful when comparing two uncertainty environments. Suppose two collateral distributions have the same expected value, but one is more damaging to the downside of the loan. Looking only at the original distributions can make the comparison hard to express, because the relevant question is not simply whether one collateral value is larger at every point. The dual distributions put both cases on the same loan-value axis and ask how solvency changes as the amount raised increases.
At a low loan value, solvency may be common under both collateral distributions. A difference in uncertainty may then have a limited pricing effect. At a high loan value, the associated promises are closer to or beyond more of the possible collateral outcomes. The difference in insolvency probabilities can become more important. This is the conceptual setting in which a surface can steepen: the pricing penalty associated with uncertainty is larger farther along the leverage axis.
The dual distribution also prepares the language used later to define stronger forms of uncertainty. Instead of saying vaguely that one distribution is “riskier,” the paper can ask whether the riskier collateral has lower solvency at every loan value, whether the ratio of solvency probabilities changes with loan value, and whether that ratio changes in the particular way that makes the credit surface steeper. The dual representation turns those questions into comparisons of default and survival along the same economic coordinate.
Pricing uncertainty is not automatically physical uncertainty
The revelation result has an important boundary. The distribution used in the basic model is the distribution that prices the collateral and bond payoffs under the assumed framework. In a complete-market interpretation, it can be understood as a risk-neutral pricing distribution. In an incomplete-market interpretation, it can represent the lender’s probability assessments. The surface therefore reveals uncertainty as it is relevant to the pricing relationship, not necessarily a single objective probability distribution observed by every participant.
Collateral prices can also contain risk premia or value derived from the asset’s usefulness as collateral. Those features do not invalidate the surface analysis, but they change how the recovered distribution should be described. It is safer to say that the surface reveals the normalized uncertainty embedded in the lender-pricing system than to say that it forecasts every future collateral outcome without qualification.
This distinction matters whenever the surface is compared across institutions or markets. Different lenders may hold different assessments of the same collateral, or they may apply different pricing conventions. The one-to-one theorem applies within the common framework specified by the model. It does not turn a market quote into an assumption-free measurement of the world.
The same caution applies to the scale-invariance result. A surface can show that normalized collateral outcomes have become more dispersed or more difficult for high-leverage promises, but it does not by itself say whether the underlying asset has become more valuable or less valuable in absolute terms. Level and shape are separate pieces of information. The model’s contribution is to make that separation explicit.
The central chain is now complete. A collateral distribution generates a loan pricing function. Inverting that function gives the promise associated with each loan value. The promise-to-loan ratio forms the loan credit surface. Because each transformation is reversible under the model’s conditions, the complete surface reveals the normalized dispersion of collateral returns. The dual distribution then expresses the same uncertainty as loan insolvency probability along the LTV axis.
That is a much richer object than one spread. But it still leaves an important question open: what should the curve look like as leverage rises? The next step is to examine its curvature and to ask why the marginal cost of leverage can accelerate, while remembering that the answer depends on the coordinate used to measure leverage.