Quick answer: securities lending is a temporary transfer of securities from a lender to a borrower against collateral, with an obligation to return equivalent securities later. The algorithmic problem has two parts. First, pricing: what lending fee or cash-collateral rebate compensates the lender for scarcity, term, counterparty, collateral, operational and recall risk? Second, allocation: when multiple borrowers want the same limited inventory, which loans should be made, in what size, and under what constraints? A strong system treats lendable inventory as a changing resource, marks collateral and exposure through time, keeps the beneficial owner’s recall rights visible, and checks whether high apparent revenue is merely compensation for concentrated or fragile risk.

A security can be worth the same market price to two investors and still have a very different lending value because one issue is abundant and the other is scarce to borrow.

Page role: this article explains the mathematics of pricing and allocating securities loans. It is distinct from securities settlement, which owns trade-to-finality mechanics, and from collateral optimisation, which owns the broader problem of assigning collateral across competing uses.

1. The basic state variables

For a given security i, define:

L_i = lendable inventory;
O_i = quantity currently on loan;
A_i = L_i − O_i = currently available inventory;
u_i = O_i/L_i = utilisation;
f_i = explicit lending fee, if applicable;
r_i = rebate rate on cash collateral, if applicable;
C_i = collateral market value;
V_i = market value of securities on loan.

Utilisation is a useful scarcity indicator. If 95% of lendable shares are already on loan, the next borrower is competing for a much smaller remaining inventory than if utilisation is 10%. But utilisation is not the whole price: a security can have high utilisation and still have low fees if supply is expected to return quickly or borrowers are not willing to pay more.

2. Borrow fees and cash-collateral rebates are two pricing conventions

In a non-cash collateral loan, the borrower can pay an explicit lending fee. A teaching approximation for annual lending revenue is:

Revenue ≈ loan market value × lending fee × day-count fraction.

In a cash-collateral structure, the borrower posts cash and receives a rebate on that cash. The lender or agent may invest the cash subject to mandate and risk controls. The economics then depend on the difference between collateral investment return and the rebate, net of fees and losses.

FINRA’s current Securities Lending and Transparency Engine rules explicitly distinguish cash-collateral loans, where rebate rates are reported, from non-cash-collateral loans, where lending fees or rates are reported. See FINRA Rule 6530 and FINRA Rule 6540.

3. Specialness: when the security itself becomes the scarce resource

Most securities can be borrowed near ordinary market terms. A hard-to-borrow or “special” security commands richer economics for the lender because demand is high relative to supply.

A simple scarcity model is:

Fee_i = base fee + scarcity premium(u_i, demand growth, available inventory) + term/risk adjustments.

The scarcity premium should usually be nonlinear. Moving from 20% to 30% utilisation need not matter much; moving from 95% to 99% can matter greatly because very little inventory remains. The exact function is a model choice and should be validated against observed loan rates rather than assumed from theory.

The New York Fed’s own securities-lending programme illustrates market-based scarcity pricing in another context: Treasury and agency loans are allocated through competitive auctions, and lending fees are applied to the market value of borrowed securities on an actual/360 basis. See New York Fed Securities Lending and its programme FAQ.

4. High short interest does not automatically imply a high lending fee

This is an important counterexample. Suppose two shares each have 10 million shares sold short. Security A has 200 million shares in stable lendable portfolios; Security B has only 12 million genuinely available to lend. Their short interest is identical, but scarcity is not.

A useful model therefore considers both demand and lendable supply:

Scarcity pressure ≈ borrow demand / effective lendable supply.

“Effective” matters because some nominal holdings may not actually be lendable due to client restrictions, voting policies, tax treatment, concentration limits, collateral terms or pending sales.

5. Allocation is a constrained optimisation problem

Suppose three borrowers request the same scarce security. Borrower A offers the highest fee but is already near a counterparty concentration limit. Borrower B offers a slightly lower fee and posts highly liquid collateral. Borrower C wants a longer term that could interfere with expected recall needs.

A simplified optimisation can be written:

Maximise Σ x_j(expected lending revenue_j − risk/capital/liquidity/operational penalties_j)

subject to:

Σx_j ≤ available inventory;
counterparty limits;
collateral eligibility and margin requirements;
beneficial-owner restrictions;
term and recall constraints;
settlement and operational capacity;
regulatory or fund-specific lending limits.

The highest nominal fee is therefore not automatically the best allocation.

6. Collateral creates a second moving value

The borrower receives a security whose price changes. The lender receives collateral whose price also changes. The system therefore has to compare the two through time.

A simple collateralisation ratio is:

m = collateral market value / loan market value.

If the agreed margin percentage is 102%, then the target is roughly C = 1.02V under the relevant valuation and convention. The New York Fed’s Treasury securities-lending programme, for example, currently specifies 102% collateral for eligible Treasury securities in its published programme terms.

When V rises or collateral C falls, margin may need to be adjusted. This reduces counterparty exposure but creates operational and liquidity flows. A static collateral snapshot is not enough.

7. Haircuts and overcollateralisation protect against close-out uncertainty

If a borrower defaults, the lender may need time to liquidate collateral and replace the missing security. During that interval prices can move. Haircuts or margin protect against this gap.

The Financial Stability Board’s securities-financing framework treats haircut methodology as an important control because too-low haircuts can amplify leverage and too-procyclical haircuts can worsen stress when they jump suddenly. See the FSB haircut framework.

The mathematical trade-off is familiar: more collateral reduces counterparty loss but increases the borrower’s funding cost and can reduce market liquidity.

8. Recalls make lendable inventory conditional

A lender may need the security back because it plans to sell, vote, meet a client redemption, change portfolio exposure or respond to another mandate constraint. Securities lending therefore is not identical to selling inventory permanently.

An allocation engine should include expected recall probability and the cost of replacing a loan that cannot be recalled smoothly. A simplified expected-value adjustment is:

Expected net revenue = fee revenue − P(recall)×expected recall/replace cost − other risk costs.

A long, high-fee loan can be inferior to a shorter, lower-fee loan if the beneficial owner is likely to need the security back.

9. Corporate actions create entitlement and voting complications

During a loan, legal title to the lent security typically transfers to the borrower, while the lender retains economic rights through contractual mechanisms such as manufactured payments. Voting rights can require recall before a record date if the beneficial owner wants to vote.

This connects directly to How Securities-Custody Algorithms Process Corporate Actions. The securities-lending engine needs the custody calendar because a profitable loan can become operationally wrong if it blocks a required election or vote.

10. Settlement fails turn lending from revenue into repair work

If the borrower fails to return the security when required, the lender can face settlement risk and replacement costs. New York Fed programme terms explicitly treat unreturned loans as fails and apply penalty economics in addition to applicable fails charges.

A robust lending system should therefore price expected fail cost, track return obligations, and distinguish a profitable loan from one whose operational reliability repeatedly creates settlement problems.

For the broader settlement state machine, see How Securities Settlement Algorithms Move Trades from Execution to Finality.

11. Central clearing can change the optimisation objective

Central clearing can replace bilateral exposures with exposures to a CCP, introduce multilateral netting and margin, and alter capital and operational costs. DTCC announced in May 2026 that the SEC had approved a new NSCC client access model for its Securities Financing Transaction Clearing Service, aimed in part at improving capital efficiency for agent stock-loan participants.

See DTCC, 13 May 2026.

The important modelling lesson is that market structure changes the cost function. A bilateral allocation rule calibrated before wider central clearing may no longer choose the same counterparties or trade sizes once netting, margin and capital treatment change.

12. Transparency changes the information set

US securities lending has historically been less transparent than many exchange-traded markets. SEC Rule 10c-1a requires reporting of covered securities-loan information to a registered national securities association and public dissemination of specified aggregated information, including loan-rate distributions.

As of August 2026, FINRA states that implementation of its Securities Lending and Transparency Engine (SLATE) has been extended to 28 September 2028. See the current FINRA SLATE timetable. This date matters because an educational article should not describe public SLATE data as if it were already live.

Once broader transaction data become available, pricing models can be challenged against market-wide rate distributions rather than relying only on bilateral histories.

13. Model inputs and outputs

Inputs	Outputs
Lendable inventory, on-loan quantity, borrow demand, fee/rebate observations	Indicative fee/rebate
Borrower limits, counterparty quality, CCP/bilateral route	Approved borrower allocation
Collateral type, haircut/margin, volatility, liquidity	Collateral requirement
Expected term, recall probability, corporate-action calendar	Term/recall adjustment
Settlement history and fail cost	Operational penalty or route choice

14. A compact allocation example

Suppose 1 million shares are available to lend. Three borrowers request:

Borrower	Requested	Fee	Risk/operational cost estimate
A	600k	6.0%	1.8%
B	500k	5.5%	0.8%
C	400k	4.8%	0.4%

Ignoring all other constraints, net economic spreads are 4.2%, 4.7% and 4.4%. The highest headline fee is not the highest net value. If B also provides the strongest collateral terms and best settlement history, the optimiser may allocate to B before A.

This example is deliberately simplified. Real systems must also account for borrower concentration, minimum trade sizes, availability fragmentation, beneficial-owner mandates, tax, recalls and legal terms.

15. Failure modes

Utilisation-only pricing. A single ratio substitutes for actual supply/demand and borrower willingness to pay.
Headline-fee maximisation. The highest fee wins even when collateral, counterparty or recall economics are worse.
Nominal-inventory error. Holdings are treated as lendable even when mandates or pending trades make them unavailable.
Static collateral. Margin is set once and not updated as either side changes value.
Recall blindness. Revenue is optimised while the beneficial owner later needs the security back.
Corporate-action blindness. Voting or election deadlines arrive while the asset is still on loan.
Fail-cost omission. Persistent bad returns look profitable because settlement repair costs sit in another system.
Transparency drift. Models are not recalibrated when new market-wide loan-rate data become available.

16. Diagnostics, alternatives and falsifiers

How does fee change as utilisation moves through 50%, 80%, 95% and 99%?
Does the model distinguish high short demand from true lendable-supply scarcity?
Which borrower produces the highest risk-adjusted revenue, not just nominal fee?
How often do recalls lead to buy-ins, substitution or failed returns?
Do collateral calls track actual market moves?
Would a simpler scarcity curve forecast observed fees as well as the complex model?
What happens to allocation when a counterparty limit tightens or clearing route changes?
Does a “special” remain special after new inventory appears?

Falsifier: suppose someone claims, “High utilisation always means a high borrow fee.” A counterexample is a security with 95% utilisation but a large amount of new lendable inventory expected tomorrow, weak incremental borrow demand and competitive lenders willing to quote low fees. Utilisation is evidence of scarcity pressure, not a law of price.

17. Verification and update triggers

Backtest quoted fees against realised accepted trades.
Measure revenue after fail, recall, collateral and capital costs.
Reconcile lendable inventory with custody and beneficial-owner restrictions.
Revalue collateral daily or at the contractually required frequency.
Track borrower concentration and settlement performance.
Recalibrate scarcity curves after major index changes, corporate actions or new supply.
Review market-structure assumptions after central-clearing changes.
Update transparency inputs when SLATE or other new reporting datasets become available.

Research anchors

The deeper lesson

Securities lending is a market for temporary scarcity. The share price tells us what ownership is worth; the lending fee tells us something different—what temporary access is worth. Utilisation, inventory, collateral, counterparty strength, recall probability and settlement reliability all change that access value. The strongest algorithm therefore does not ask only, “What fee can we charge?” It asks, “Which loan creates the best risk-adjusted use of a scarce, recallable asset, and what evidence would show that our scarcity model has stopped matching the market?”

Educational note: This article explains public securities-finance mathematics. It is not short-selling advice, investment advice, a securities-lending recommendation or an operational guide for any institution.

Small Group Tutorials

How Securities-Lending Algorithms Price and Allocate Stock Loans: Borrow Fees, Utilisation, Specials, Collateral, Recalls and Fails