Valuation of Illiquid Loans Beyond Gut and Guesswork with AI

Article 1 in a series of 3

• Link to second article: Quantifying Illiquidity from Theory to Reality

• Link to third article: Can AI Solve the Puzzle?

Introduction

With the growth of private credit, the valuation of illiquid loans has become more and more important, as banks, insurers and pension funds are increasing their allocations to private credit and therefore illiquid loans.

This is the first article in a series of three short articles on the valuation of illiquid loans. In this series we will be covering:

• Why valuation is important,

• What the regulations say,

• How illiquidity can be measured,

• Which valuation models are available,

• What the data challenges are,

• How the data challenges can be measured and

• How AI can help address these challenges.

We have been closely involved in developing and implementing a solution for valuing illiquid loans, and we will draw on this hands-on experience throughout these articles. In addition to sharing practical examples, we aim to take the explanations beyond our direct work, offering a broader perspective on the challenges and solutions to value illiquid loans.

Why Illiquid Loan Valuation Matters

When markets are liquid and data is abundant, valuations are relatively simple. A look at a Bloomberg terminal, or a subscription at a data vendor is usually enough to determine the value of a financial instrument. This is, however, completely different for illiquid loans. There are no public marketplaces where loans are trading and data is mostly private and difficult to come by.

For banks, the valuations of illiquid loans are important for their trading and syndication books, which are measured at Fair Value. Additionally, the Net Asset Valuation (NAV) financing that banks provide to funds depends on the valuations of the underlying loans. For insurers, illiquid loans as an investment need to be Fair Valued for Solvency II calculations.

Valuation is not just an accounting formality, as it hits the P&L, capital and leverage ratios of banks and insurers. Also, for pension funds, the valuation immediately affects the estimated payouts to beneficiaries.

Small differences in the assumptions can have large effects on the reported results. That is also why regulators take a keen interest, as they have an interest in the valuations, to track the financial health of parties active in the Financial Markets and Private Credit market in particular.

The Valuation Framework: IFRS and Fair Value Levels

IFRS defines Fair Value as the price that would be received to sell an asset, or paid to transfer a liability in an orderly transaction between market participants at the measurement date.

This sounds simple, but IFRS is principle-based and not rule based and does not prescribe how the Fair Value should be assessed or calculated in detail. IFRS defines a hierarchy of three levels, which classify how Fair Value is measured based on the observability of inputs. See Figure 1 below.

Figure 1: IFRS valuation hierarchy

For illiquid loans, quoted market prices are difficult to come by, and finding comparable instruments is also a challenge. So, at best, illiquid loans can be classified as level 2, but will often need to be classified as level 3, as the valuation of illiquid loans often depends on models, indirect data and can be more art than science.

(Il)liquidity is not Binary

Liquidity is not black and white, it exists on a spectrum. Key factors that influence liquidity are shown in Figure 2. The combination of these factors determines how liquid a loan is.

Also, loans may be liquid shortly after closing in the primary market, until all market participants have their desired allocation, after which liquidity dries up. Market sentiment is also an important driver of liquidity. In stable times there is usually more liquidity than during periods of stress.

Figure 2: Liquidity Drivers

What makes Loans different from Bonds

Loans and bonds have both interest and repayments, and that’s where the similarity ends. Bonds are standardized and tradable securities that settle on exchanges/clearing houses. Loans, however, are privately negotiated, and mostly tailor-made instruments and they settle through an agency construction.

All differences between loans and bonds originate through these differences and it makes loans inherently more complex than bonds. Figure 3 graphically illustrates some of these differences:

• Different loan products: Loans can be term loans with pre-agreed repayment schedules, or they can be of a revolving nature, where the borrower can freely draw and repay.

• Multiple Pricing Elements: Loans are mostly floating rate, while bonds are mostly fixed rate instruments.

• The interest margin of loans may differ over time or may be dependent on borrower performance. Next to interest, different types of fees, like commitment, or utilization fees can be part of the loan agreement.

• Embedded options: Loans usually have embedded options, like prepayment, delayed drawdown, cancelation, or extension. Also, currency switching, or interest period switching can be embedded.

• Collateral and guarantees: loans may have collateral as a cover, or guarantees by parent companies, or third parties to mitigate risks.

Figure 3: Loan Cash Flows

The Valuation Process

The valuation process consists of three steps, as also shown in Figure 4:

• Data gathering, which consist of two sub-steps:

  • Collecting all relevant financial information, such as loan and borrower details

  • Collecting all relevant market data, such as various spreads, any pricing information about previous transactions, but also yield curves, FX rates, etc

• The calculations itself, which can also be divided into two sub-steps:

  • Projections of expected cash-flow, including client behaviour, such as prepayments and/or drawing of additional funds, if applicable

  • Discounting of the projected cash flows against the appropriate rate

• Validation of the calculation results, by monitoring the accuracy and behavior of the model, price testing by experts, etc.

Figure 4: The valuation process

How AI can contribute to the Valuation Challenge

Valuing illiquid loans requires expert judgement, complex data synthesis, and model-based assumptions. Artificial intelligence (AI) can support this process, not by replacing human expertise, but by enhancing the information on which decisions are based and by improving the consistency, speed, and transparency of valuation workflows. Note that human judgement and oversight remains important when critical decisions are being made by AI systems

During the data gathering phase, AI can contribute to four main areas:

• Automating Data Extraction: Large Language Models (LLMs) and document intelligence systems can read and parse loan agreements, term sheets and covenant documents. This enables automatic extraction and structuring of key variables such as maturity, pricing grids, or collateral terms, reducing manual effort and transcription risk.

• Enhancing Data Enrichment: AI can cross-reference internal and external datasets (e.g. transaction databases, credit ratings, market indices) to fill data gaps and improve the representativeness of the input data. This allows better calibration of model parameters such as spreads or liquidity discounts.

• Clarifying and Classifying Loan Attributes: Machine learning models can automatically classify loans by borrower type, industry, geography, or liquidity features (e.g. currency and deal size). This structured tagging enables more granular analysis and more accurate model segmentation, as well as explainability.

• Scoring Relative Liquidity: By integrating textual data (loan terms) with quantitative inputs (secondary prices, deal activity, macro factors), AI models can estimate the relative liquidity score for individual loans. This helps identify which loans are likely to be harder to exit, or which segments may experience market stress earlier.

These contributions do not automate valuation outcomes. Rather, they make the process more evidence-based, repeatable and auditable. Expert oversight remains essential, AI is only an assistant that supplies cleaner, richer data and also an extra eye that can highlight relationships that may otherwise remain hidden. The final step in the valuation process will therefore need to cater for managing AI risk, such as hallucination, model risk and bias. Keeping track of logs, reviews and overwrites during the full valuation process are also essential for model monitoring and governance.

In the next article we will elaborate on steps 2, 3 and 4, as indicated above. We will focus on quantifying illiquidity and calculating Fair Value for individual loans. The approach uses data from CDS, bond, and loan markets to estimate spreads. We will address the practical data hurdles (coverage gaps, inconsistent terms and entity resolution) and show how AI can help to overcome these challenges.

Acknowledgement

We would like to thank S&P Global Market Intelligence for making available the loan, bond and CDS data that we have used in this series of articles.

Disclaimer

The views shared in this article are our own and don’t represent the views of our employer.