KEY CONSIDERATIONS FOR CREDIT RISK ASSESSMENT DURING CRISIS
Introduction
The credit departments and functions of financial institutions and other credit-sensitive organisations will need to burn the midnight oil during the pandemic (albeit hopefully from home-offices rather than in crowded corporate offices). From a credit perspective, lenders should be more prepared and more capable than was the case during the financial crisis that commenced in 2007/08.
High-Level Actions and Considerations
There remain far too many unknowns about the virus to reliably predict how the human impact (in terms of life and health) will evolve during the pandemic. However, high- level economic scenarios are somewhat easier to establish on geographical, and industry sector bases albeit such scenarios require regular surveillance and update particularly as State (or sub-State) support measures are contemplated or implemented.
A satisfactory credit risk framework already includes procedures to be implemented during a crisis (and a good framework enables rapid enactment of those procedures into speedy results)
Lenders will partition their portfolios into those expected to be most adversely affected by the crisis initially as a function of quantum of credit exposure, geography, industry and existing credit rating/score pre-crisis. This prioritises which exposures are the more urgent ones to review.
Sub-portfolio would then make refinement of those partitions by reference to the results of initial high-level stresses, consideration of non-financial risk factors (e.g. management and operations) and any known developments since the last credit assessment.
This finalises the priority-ranking (for review) of the existing portfolio.
Individual borrowers (or broadly homogeneous sub-groups) are then examined in more depth. The focus of that examination is not simply to determine a new rating/score but also to proactively optimise strategy for each borrower (or groups of borrowers).
In particular, consideration will be given to the merits of providing increased or additional facilities to specific borrowers (and perhaps reducing facilities for others); whether facilities should be proactively restructured in order to better align them to borrowers (changed) needs while minimising any potential economic loss for the lender; and the strategic, financial, business and reputational pros and cons of pursuing various courses of action.
The analysis and results will also inform and dictate how applications from potential new customers will be assessed.
Key Considerations
Some loans or facilities are reliant (for repayment) on the borrower successfully refinancing the loan. Examples include short-term bridge financing to acquire properties or for real estate investment projects (in advance of longer-term finance).
Merton Model Type. A number of credit assessment tools deploy a Merton-related approach for assessing PD (probability of default). The Merton approach (in very broad terms) estimates the likelihood of the value of assets falling below the value of liabilities as an anchor point for then determining PD.
In practice, borrowers avoid (actual) default from this anchor point if their cash-flow (plus available liquid assets) are such that they can continue to service their liabilities as they fall due in a timely manner. Equally, borrowers that have a healthy excess of assets over liabilities will default if they are not sufficiently cash-generative (and are unable to raise short-term finance on sensible terms) to enable them to service their obligations as they fall due.
Despite these flaws – and no approach is without flaws – using Merton model approaches (appropriately) is fine in most economic circumstances.
However, in crises such as this one (and as was the case during the financial crisis) the Merton approach will become less reliable for PD assessment since (a) a firm’s liquidity position will drive default, and (b) the market conditions for liquidity (supply/availability and cost of short-term funding) are likely to be very different to the norm. Additional scrutiny and analysis is needed if your core model is Merton-based.
The Merton approach remains valid for assisting with the assessment of LGD (loss in the event that default happens) although significant (analytically consistent) revisions to assumptions are required for estimating LGD under realistic, hypothetical stress scenarios.
In fact, the strength of a Merton-based LGD approach in the current crisis is more about driving strategy and action than individual LGD estimates themselves. In particular, it can assist in determining whether a lender should proactively restructure existing facilities for a borrower in advance – to partially mitigate stress for the borrower, reduce PD risk and optimise (or hopefully eliminate) economic loss.
Market-Driven Approaches. Approaches that rely on movements in quoted financial instruments and quoted assets (e.g. bond prices, equity prices, CDS spreads, etc) are often quick to react to new market information and – even when not deployed as a core model for estimating PD – have the potential to raise timely action flags to review a particular credit when those indicators signal significant deviations from a core model score.
By their nature, market-driven scores can be volatile, can suffer from incomplete (or misunderstood information affecting market prices) and have other drawbacks. Equally, they can form an integral and valuable component of a good, comprehensive risk assessment framework as part of the “checks and balances” on core model results.
However, during the current crisis increased volatility is inevitable (as matters evolve) and the “signals” provided are going to be based on less complete information than usual. Therefore, while such signals should not be ignored, they should be treated with greater caution than normal.
Machine Learning Type Approaches. Some new entrants have explored deploying machine learning techniques to develop credit assessment models over the course of time.
Unfortunately, a number of such entrants have chosen a strategy of “reinventing the wheel” rather than simply starting with the existing wheel and then leveraging the technology at their disposal to try and refine and improve it.
The result is that (while the model can get better with each iteration) such models often take several years to reach a level that would be deemed “satisfactory”. Models (reliant solely on machine learning) at a barely satisfactory level are likely ill-equipped to cope with credit assessment during the crisis: the models will (of course) “learn” further from what happens during the crisis, but that learning process may incur a high cost.
(Past) Financial Data. Any assumption that past financial profiles are broadly indicative of a firm’s prospective financial profile over the short to medium term is no longer valid for many (if not all) firms. There is a need to overlay (or stress those) financials with the expected impact of the crisis on demand, supply, income and outgo. As this is an evolving situation, several prospective horizons are needed for the adjusted financial position.
The current financial profile (and how that evolves over the months ahead) will only be available to you next year (when 2020 accounts are published). In essence, it may look very different from the latest available report and accounts.
Even for specialised finance classes, when the prospective analytical focus is (arguably) more indepth because of the lack of pertinent past financial information, standard forward-looking assumptions related to the demand, supply, timing, and cash-flow need to be similarly revisited and revised in the context of the crisis)
Financial Benchmarks. These require (analytically consistent) revision. Conceptually, scoring benchmarks (whether continuous, discrete or otherwise) reflect the required value of a financial metric associated with withstanding a particular magnitude of stress (whether that stress relates to the economy, the sector, is idiosyncratic to the firm or is a combination of economy, sector and firm-specific attributes).
Over the course of a cycle, benchmarks tend to be generally invariant (with the exception of amount metrics) unless the industry concerned is a volatile one or exhibiting a structural change. Even for volatile sectors, the benchmarks for metrics at the “bottom” - when differentiating between the worse credit risks - remain fairly static. For differentiating at the “top” the value of the metric “relative to peers” tends to differentiate better so those scoring benchmarks may change more frequently.
However, in the current environment, the focus is on what the financial profile looks like after significant stress has already happened. Revised scoring benchmarks, therefore, need to be set at levels that are commensurate with the (changed) probabilities of further stresses of associated magnitude from that point.
Non-Financial Risk Factors. These risk factors require rescoring. Much greater consideration is needed (during the pandemic) in particular on aspects of management, strategy and how the business is being operated.
Different time horizons also need consideration. Actions already are taken (or not taken) to partially alleviate short-term adverse pressures may be credit-positive over shorter horizons but rebound (i.e. be credit-negative) over the medium term. In terms of firms exhibiting similar stressed financial profiles, management and operational differences may significantly differentiate credit quality between them over all time horizons.
External Support Factors. Naïve, simplistic approaches to factoring in a group or parental support are likely insufficient during the crisis, particularly if the parent or group itself also suffers a deterioration in its overall financial profile.
To accurately evaluate the impact of group support it will be important to move beyond the group’s (stressed) “financial capacity” to provide support to subsidiaries and associates and focus strongly on also measuring the group’s willingness to provide support if needed and the economic and strategic rationale for rendering any support.
State Measures. Most governments are taking or considering taking measures to assist businesses and consumers during the crisis. In general, such measures are effectively direct or indirect liquidity support that will dampen the economic severity of the crisis and/or spread that impact over time.
The effect of these measures on a firm’s credit quality need to be incorporated not only into its (stressed) financial profile over each horizon but also integrated into the determination of revised benchmarks for financial metrics.
Loss Given Default (LGD) – Prior to Default. Assumptions for the determination of EVA (the economic value of assets) and the statistical distribution of EVA (primarily the volatility factors) require revision in an analytically consistent manner with the scenarios being considered in the PD context. However, these should flow naturally as a by-product of the PD analysis. The stress level to be applied to estimated (stressed) EVA should also be consistent with the severity of the stress scenario.
Consideration also needs to be given to whether jurisdictions will become temporarily less creditor friendly during the crisis period; any changes in the costs associated with realising collateral; and changes in the timing for the realisation of collateral.
In fact, LGD assessment in the current crisis may be more about driving strategy and action than individual LGD estimates themselves. In particular, it can assist in determining whether a lender should proactively restructure existing facilities for a borrower in advance – to partially mitigate stress for the borrower, reduce PD risk and optimise (or hopefully eliminate) economic loss.
Loss Given Default (LGD) – Post-Default. Assumptions for liquidation values by type of asset require revision in an analytically consistent manner with the scenarios being considered in the PD context. In particular, if the scenario infers a significant increase in defaults among firms with similar assets, the realisable value of specific asset types will naturally be significantly depressed.
How Ready Are Lenders?
From a credit perspective, lenders should be more prepared and more capable than was the case during the financial crisis that commenced in 2007/08.
At that time, organisations were still preparing for Basel II, Solvency II and other regulations. The financial crisis revealed that many methodologies (then in place) were not “fit for purpose” and led to improvement. Subsequent regulatory requirements migrated away from the initial “concept” of Basel towards led to a more proscribed and (in some cases) prudent but onerous regulatory framework, increased focus on stress testing, and IFRS9.
Offsetting this, the change to LRA (long-run average) default definitions from TTC (through-thecycle) definitions means that some aspects of analysis (that were challenging for lenders) have been consigned to the “back burner” and a greater reliance is now often placed on using historical data in an almost mechanical manner without appropriate prior adjustment (following its analysis) – rendering it (in some situations) less representative of the risks that are to be assessed.
As memories of the financial crisis receded (some) lenders have unfortunately reverted to a “penny wise and pound foolish” approach for their credit framework while not recalling that “you only get what you pay for”. Also, some new (challenger) institutions have indulged themselves by seeking to reinvent the wheel rather than improving the wheel that already existed.
However, on balance, many credit-sensitive institutions should be better prepared.
Conclusion
Effectively managing credit during a crisis will require significant additional analysis, modelling and resources but (in general) lenders are better placed to do so than during the 2007/08 financial crisis.
Contact us
EMEA
Risk-Enterprise Limited
Email: info@risk-enterprise.com