No model and no loss function is not a recipe for good macroprudential policy

Author: Richard Barwell

Published: 27 September 2021

There is a problem with macroprudential policy. The two core building blocks of an economic policy regime – a model of the system and a loss function to guide policy decisions – are missing[i].  Without a reliable model of the system, policymakers cannot forecast with any accuracy how that system will behave in the future or how it will respond to policy interventions. Without a loss function, policymakers are unable to evaluate outcomes and hence whether they are making the situation better or worse.

Incomplete foundations

The ideal model for policy purposes captures all the salient features of the system that policymakers are trying to influence.  In the macroprudential sphere, that model would ideally provide a compelling and internally consistent description of:

  • the key features of the system that ultimately contribute to the financial cycle, such as the build-up of leverage, debt, liquidity, maturity mismatch and bubbles within the financial sector; and
  • the macro interactions between the financial sector and the real economy that are reflected in the financial positions of the household and corporate sectors and in particular the state of the property market; and
  • the institutional detail on both sides of the regulatory perimeter within the financial system, such as the state and structure of the balance sheets of systemically important institutions, and how the complex web of interconnections between institutions and across markets evolves in response to market conditions and policy interventions; and
  • the correlated shifts in the behaviour and beliefs of chief executives and risk-takers at major financial institutions and within the broad investment community that drive market outcomes; and
  • how the policy levers at the policymaker’s disposal influence those system features, macro interactions, institutional detail and behaviours and beliefs.

The loss function codifies the objectives and preferences that the social planner would like the policymaker to adopt by translating outcomes on a relevant set of variables into a measure of social welfare, typically defined as a loss relative to some ideal level. The policy problem is then transformed into choosing the policy setting that minimises the social loss, given the policymaker’s understanding of and uncertainty about the current state and structure of the system. Without a working understanding of that loss function, it is hard to know how to calibrate the policy stance (you can observe outcomes but not rank them) and even whether you are contravening the Hippocratic Oath of economic policy: first, do no harm.

At this point, macroprudential policymakers may protest that all policy regimes – not just macropru – fall short of this demanding standard of a well-articulated loss function and a fit-for-policy-purpose model of the system.

One way to think about the extent of the missing model and missing loss function problem in macropru is to compare the situation with monetary policy.  As a former Governor of the Bank of England observed, the practice of monetary policy has at times been ahead of the theory[ii], and the latest generation of models used in the current monetary policy debate are clearly far from perfect. Nonetheless, it seems uncontroversial to argue that the state of knowledge is far more advanced (and the problem far less complex) in the monetary sphere than in the macroprudential sphere. Likewise, whilst there is a broad consensus among policymakers and academics over the loss function of monetary policy, significant disagreement remains over the macroprudential loss function.

Thoughts on the loss function – stability

Making economic policy is hard enough even with a clear expression of your objectives, given the inevitable limitations of the model of the economic system that policymakers are trying to influence. But life is much more complicated without a target to aim for. The basic objective is clear: to reduce the frequency and severity of episodes of financial instability. But in order to make progress towards a loss function, we need a working definition of financial instability and the macroprudential agenda, and this is where the disagreement begins.

The orthodox definition of financial instability focuses on the resilience of the financial system. Before the crisis, a distinction was made between a microprudential focus on the resilience of individual institutions and a macroprudential focus on the resilience of the entire system to capture the externalities and coordination problems that can be lost in a ‘one institution at a time’ approach[iii].

Since the crisis, the emphasis has shifted a little, with a greater focus on near-misses as well as full-blown crises.  The social planner may conclude that financial stability has not been preserved if the financial system withstands a moment of stress but only by taking defensive actions – in particular, curtailing the provision of critical services – that have a material negative impact on the real economy[iv].

For many, the macroprudential perspective on financial instability now includes an explicit focus on safeguarding the provision of core financial services. Of course, numerous financial services are provided across multiple markets by a large set of institutions to a diverse range of customers, and there is no single supply schedule across all those markets. Nor is there a clear definition of which services should be considered core, nor what constitutes a sufficiently severe disruption.

Heterodox definitions of financial instability suggest more ambitious visions for macroprudential policy. Rather than just seeking to avoid sudden stops in the provision of core financial services, there is an argument that macroprudential policy should seek to ‘smooth the financial cycle’[v].  The regime is now less asymmetric and seeks to stabilise the provision of those services (i.e., leaning against both inwards and outwards shifts in the credit supply schedule that are not warranted by fundamentals) or even credit flows themselves.  More ambitious still, policymakers could seek to influence asset prices, or at the very least property prices[vi].  Most ambitious of all, financial instability could be defined as “any deviation from the optimal saving-investment plan of an economy that is due to imperfections in the financial sector”[vii].

These alternative expressions of the stability objective correspond to radically different trajectories for the economy, which then require radically different visions for the scale and nature of macroprudential policy interventions. At one end of the spectrum, one can envisage an economy in which there is still significant volatility in credit flows, debt stocks and asset prices, but there is now sufficient resilience within the banking system to prevent, say, a crash in the housing market from bringing down the banks.  Increasing ambition implies more significant intervention to damp the cycle in credit flows and then risk premia. Finally, at the other end of the spectrum, there are interventions designed to mitigate any friction within the system that impedes the first-best allocation of risk and resources.

Thoughts on the loss function – introducing equity and efficiency

The focus on stability is understandable, but efficiency considerations should and equity considerations could also feature in the macroprudential loss function. Otherwise, policymakers will be obliged to act like “resilience nutters” and deliver an economy that is too stable.

The debate about whether you can have too much stability typically focuses on the implications for efficiency. There is a risk that macroprudential interventions might indirectly constrain the flow of funds to the SMEs that ultimately drive productivity growth. Concentrated oligopolistic banking systems may also be more resilient than competitive systems, but again there is a loss of efficiency involved. However, it might also be the case that resources are misallocated in credit booms, and the supply side may suffer lasting damage in a credit bust. In other words, macroprudential interventions may have a positive impact on efficiency too.

The equity dimension of macroprudential policy is also important. A “back to the future” economy in which prudent banks only lend funds to low-risk individuals is likely to be one in which more low-income households will face severe credit constraints. The impact on aggregate consumption may be modest, but the impact on the welfare of those concerned would not be[viii].

More broadly, relatively few macroprudential interventions are likely to prove pareto improving – there are inevitable distributional consequences from any intervention that influences asset prices. Macroprudential policymakers could choose to ignore the distributional consequences of their actions and leave the fiscal authority to manage the fall-out – although it is debatable whether fiscal policy can easily offset a structural, selective tightening in credit conditions. What is clear is that if the macroprudential regime does ignore the equity dimension, then it is imperative that this is understood and supported by the political class.

Translating thoughts into an actionable loss function

One common feature across the various elements of the hypothetical macroprudential loss function is the reliance on concepts that are either inherently model-based or hard to measure in practice. Where stability is concerned, the starting point is the will-o’-the-wisp concept of systemic risk (that loosely captures the likelihood and severity of a crisis event occurring that has material macroeconomic consequences), but the macroprudential authority also has a specific interest in the location of the supply schedule for specific financial services across multiple markets in future states of nature. On the efficiency front, the focus is on identifying the impact of developments within the financial sector on the level and future growth of potential supply via a number of potential theoretical channels. On the equity front, the focus is on quantifying the extent of credit rationing within segments of society and estimating the consequences for welfare.

Another key issue is the dimensionality of the macroprudential problem. The policymaker requires some guidance on the relative importance of each of the many arguments of the loss function: the disruptions across multiple credit markets or outcomes across regions or income groups within society.

One way to calibrate the importance of disruptions in each credit market is according to the consequences for aggregate activity – the so-called GDP at risk approach[ix]. If your perspective is macro stabilisation, the GDP at risk approach makes a lot of sense but it has implications. First, the regime will be relatively insensitive to disruptions that disproportionately impact marginal sectors of the economy or sections of society that make a modest contribution to aggregate output. Second, the calibration of the macroprudential regime then hinges on assumptions about the capacity of the fiscal and monetary authorities to manage the decline in output that would follow a future financial crisis.

The issue of how to handle winners and losers on the equity front is at least a familiar public policy problem, although probably not one the macroprudential authority is keen to grapple with. The specific issue here is how to protect the interests of the least well off in society who are both likely to be under-represented in the GDP-at-risk framework and not necessarily well served by the subset of policy interventions that make the financial sector more resilient by effectively prohibiting lending to very high-risk individuals.

The over-arching problem here is the absence of a reliable model of the system the macroprudential authority is trying to influence. In principle, the precise calibration of the optimal loss function in any regime should be model-consistent – in other words, objectives are endogenous, not exogenous[x]. By a stroke of luck, it can be demonstrated that the quadratic loss function that has been adopted by convention in the monetary realm may be a reasonable approximation to the model-consistent loss function that emerges in modern macro models, where that loss function reflects the welfare of the representative agent[xi]. Unfortunately, it is not obvious that future research will validate any ad-hoc macroprudential loss function that might be adopted today in the same way.

The macroprudential loss function in practice

The current cohort of macroprudential policymakers are not without guidance. Here in the UK, the objectives of the Financial Policy Committee (FPC) are established in the Bank of England Act and then clarified in regular reviews by the Chancellor.

The Committee’s primary objective is to protect and enhance the resilience of the UK financial system, and subject to that, the FPC has a secondary objective to support the economic policy of the UK government.  The FPC is supposed to deliver that primary objective by identifying, monitoring and then taking action to remedy “systemic risks”. The FPC is instructed to consider risks to a “significant part” of the UK financial system, and not just the entirety, and reminded that the purpose of resilience is to contribute to avoiding “serious interruptions” in the provision of “vital functions” by the financial system.

The efficiency dimension of macroprudential policy is recognised in the Committee’s remit. There is a reminder that the Government has a constant interest in growth and employment. The Act also makes clear that the FPC is neither required nor authorised to take action that it believes would likely have “a significant adverse effect on the capacity of the financial sector to contribute to the growth of the UK economy in the medium or long term”.

The equity dimension is less apparent in the guidance provided by the Government. The Committee is informed that the Government’s objective is to achieve “strong, sustainable and balanced growth” including levelling up across the regions, as well as maintaining a financial system that provides productive finance and critical financial services.

There is a conscious effort here to steer policy but the guidance falls short of an explicit loss function that specifies precisely what the FPC is trying to achieve.

It is not entirely clear where the FPC should position itself on the spectrum of resilience objectives, between the orthodox and heterodox visions outlined above. Nor is it entirely clear how the Committee can be confident that it will comply with the Government’s instructions without a fit-for-policy-purpose model. For example, it is unclear how the Committee can be confident that it will not inadvertently have a significant adverse effect on trend growth without a model that can be reliably used to quantify the impact of policy interventions on the supply side of the economy (it is also unclear what constitutes a “significant adverse effect” in the context of trend growth).

Where possible, the FPC is supposed to conduct cost-benefit analysis of each policy intervention but that exercise may be likewise compromised by the missing model and loss function. Not only does the macroprudential authority need to be able to accurately forecast outcomes and know how to value them in order to complete that exercise, she also needs to know the scope of that exercise – in particular, are distributional considerations relevant and if so, how?

Coordination problems and the über policymaker

Macroprudential policy is not the only game in town. There are multiple policymakers that have a significant interest in, or whose actions are likely to have a significant impact on, the financial stability agenda. Unfortunately, the absence of clear loss functions exacerbates the coordination problem that naturally arises in this situation.

There is a microprudential supervisor of core financial institutions, distinct from the macroprudential authority. There is a financial conduct authority, a resolution authority and a regulator of core financial market infrastructure that each have a direct interest in key parts of the financial instability jigsaw. There is a central bank that sets haircuts against collateral and can act as a lender and market maker of last resort in a crisis. Then there are the monetary and fiscal authorities: each responsible for an aspect of economic stability. There is even a competition authority that has an interest in the provision of core financial services.

There is no neat separation of responsibilities. We can distinguish between ‘dynamic policies’ and ‘structural time-invariant policies’[xii] in theory but we cannot treat the microprudential regime as an inert, acyclical regulatory floor in practice. Nor can we assume that the interests of the macroprudential authority will always coincide with those of other policymakers. There may be a tension between safeguarding the resilience of financial institutions and safeguarding the provision of core financial services in moments of stress. There may be moments when the monetary authority believes an increase in the level of household debt is a price worth paying to achieve price stability in the short-term but the macroprudential authority disagrees and believes that an increase in debt threatens financial stability in the long run.

Practical solutions to this problem lie in a fully-specified financial stability policy framework that would articulate specific responsibilities to each of these policy institutions with a stake in financial stability policy, allocate instruments and loss functions accordingly and then delivers coordination between those regimes via memoranda of understanding. The further we depart from that fully-specified framework, the more ambiguity creeps into this financial stability regime of multiple policymakers.

To some extent, the coordination problem is managed in the United Kingdom by virtue of the fact that the Governor of the Bank of England is the key decision-maker in many of these institutions.  Lack of clarity on the loss functions across those institutions, therefore, vests considerable power in his or her hands.

Accountability deficit and framework hysteresis

Critically, if policymakers have no clear guidance on how to evaluate outcomes, no clear target to aim for and no precise guidance on the scope of the regime then it is hard to see how the media or the political class can provide effective scrutiny of the decisions taken by unelected officials.  In other words, the missing model and loss function can undermine accountability too.

To make matters worse, it seems reasonable to suppose that the audience is both uninformed and inattentive and that further undermines accountability. The subject matter is complex and inaccessible. Shadow banking is not a subject that naturally lends itself to stories on the front pages of the newspapers (until it is too late) and is therefore perhaps not the obvious focus of parliamentary scrutiny. There is no data series published by the statisticians which can be used to benchmark the performance of the macroprudential regime in the same way that the inflation data can be used as a cross-check on the monetary authority.  Society is painfully aware of moments of financial instability but they are rare and in the long booms between the busts there is not always a consensus among the experts that the risks to financial stability are rising so the focus of the public debate on economic policy naturally shifts away.

Ambiguity over the loss function – the objectives and scope of policy – coupled with this limited accountability could give rise to framework hysteresis. The regime may ultimately become defined by precedent in a world where nobody is quite sure what the regime is precisely for. Macroprudential policy will become what previous macroprudential policymakers did, and for the reasons they did it. One interesting possibility here is a tendency towards macroprudential mission creep. If policymakers are both uncertain about their capacity to detect and prevent future crises and nervous about deviating from precedent then a ratchet effect may arise in which innovation interventions today, based on an expansive interpretation of the remit, could reset expectations of what the regime is for tomorrow.

Conclusion

The lack of a model and a loss function wasn’t a major problem back in 2010 because it was obvious to almost everyone what had to be done and how to do it. Policymakers needed to build resilience within the financial system as a matter of urgency. But now that the post-crisis reform agenda is approaching completion the case for completing the foundations of the macroprudential framework is compelling.

No model and no loss function is not a recipe for good policy. Unfortunately, filling in the blanks is a formidable task. The only solution is (to paraphrase Churchill) blood, toil, tears and sweat.  Years (and most likely decades) of academic research into macroprudential policy should clarify our collective understanding of the nature of the problem and improve our capacity to measure the relevant features of the system.  Eventually, we will learn what really matters, how to value it and how to measure it.  A target – analogous to 2% CPI inflation – might eventually emerge. In the meantime, policymakers will have to do without and that suggests as much clarity as possible from all concerned on what the regime is for and why, and then making sure policy actions and communications are consistent with those marching orders.

Disclaimer: The content of this blog represents the views of the author and not his employer.

Endnotes

[i] For more detailed exposition of the argument in this article see Barwell, R (2013), Macroprudential Policy and Barwell, R (2017), Macroeconomic Policy after the Crash (Part ii): Issues in Microprudential and Macroprudential Policy.

[ii] King, M (2005), ‘Practice ahead of theory’, Speech.

[iii] Crockett, A (2000), ‘Marrying the micro- and macro-prudential dimensions of financial stability’, Remarks.

[iv] Rosengren, E (2011), ‘Defining Financial Stability, and Some Policy Implications of Applying the Definition’, Keynote Remarks.

[v] Constâncio, V et. al. (2019), Macroprudential policy at the ECB, Occasional Paper.

[vi] Constâncio, V (2016), ‘Principles of macroprudential policy’, Speech.

[vii] Haldane, A, Saporta, V, Hall, S and Tanaka, M (2004), ‘Financial stability and macroeconomic models’, Financial Stability Review.

[viii] Crossley, T and Low, H (2014), ‘Job loss, credit constraints and consumption growth’, The Review of Economics and Statistics.

[ix] See, for example, Adrian, T, Boyarchenko, N and Giannone, D (2019), ‘Vulnerable Growth’, American Economic Review.

[x] Walsh, C (2005), ‘Endogenous objectives and the evaluation of targeting rules’, Journal of Monetary Economics.

[xi] Woodford, M (2003), Interest and prices.

[xii] Liang, N (2017), Rethinking financial stability and macroprudential policy, Brookings.

Author

  • Richard Barwell

    Head of Macro Research at BNP Paribas Asset Management. Richard is responsible for promoting collaboration between investment teams and formulating alpha-generating investment views across all asset classes. Richard has worked at Royal Bank of Scotland (Markets & International Banking and Global Banking & Markets) and the Bank of England as a Senior Economist. Richard has 16 years of investment experience.