class: center, middle ## Prudential fiscal stimulus MMF Conference, 6 September 2022
#### Alfred Duncan #### University of Kent
#### Charles Nolan #### University of Glasgow
This project is supported by research funding from UKRI Grant Ref: ES/V015559/1 --- class: left, middle ### The problem (the Greenspan put)
Stimulus policies can increase moral hazard.
Anticipating to be rescued in downturns, firms might take more risk today.
This paper asks: - Can we stimulate the economy in a way that encourages prudence?
--- class: left, middle ### Our contribution
- We show that countercyclical wage subsidies can improve welfare, even in the absence of aggregate demand externalities. - We characterise the optimal wage subsidies. - We estimate that simple rule implementations can generate large welfare gains.
Figure: Countries with new or existing wage subsidy schemes during the Covid-19 pandemic (Sources: ILO, IMF, authors’ calculations)
--- class: left, middle ### Intuition behind our result Absent intervention
--- class: left, middle ### Intuition behind our result Introduction of the wage subsidy
--- class: left, middle ### Intuition behind our result Firm's risk allocation response
--- class: left, middle ### Intuition behind our result
--- class: left, middle ### Intuition behind our result
Labour supply is a complement to firms' inside wealth.
Anticipating wage subsidies in recessions, firms will be more prudent in expansions.
The ex-post wage subsidy replicates an ex-ante macroprudential intervention.
--- class: left, middle ### Related literature
Macropru - di Tella (2017); Duncan and Nolan (2022). - Farhi and Werning (2016); Schmitt-Grohe and Uribe (2012); Information economics - Arnott and Stiglitz (1991)
--- class: left, middle # The macroprudential externality --- class: left, middle # The macroprudential externality ## Three ingredients 1. Anonymous unrestricted trade in aggregate-state contingent securities. 2. Agency costs restricting trade in idiosyncratic-state contingent securities. 3. Risk aversion over individual specific states.
Jump to results
--- class: left, middle ### The entrepreneur's intratemporal problem
$$ R(\theta,s),b^e,h^e =\arg \max \mathbb{E} \left\lbrace \hat{v}^e(R(\theta,s)q^e) | \theta\right\rbrace $$ subject to $$ R(\theta,s)q^e \leq f(\theta,k,h) - r^b(\theta)b^e - wh^e\qquad \forall \theta$$ in addition to $$ [\text{capital budget constraint}] $$ $$ [\text{truth telling constraint}] $$ $$ [\text{lender participation constraint}] $$
--- class: left, middle ### The entrepreneur's intratemporal problem
$$ R(\theta,s),b^e,h^e =\arg \max \mathbb{E} \left\lbrace {\color[rgb]{0.7372549019607844,0.5882352941176471, 0.9019607843137255}\hat{v}^e}(R(\theta,s)q^e) | \theta\right\rbrace $$ subject to $$ R(\theta,s)q^e \leq f(\theta,k,h) - r^b(\theta)b^e - wh^e\qquad \forall \theta$$ in addition to $$ [\text{capital budget constraint}] $$ $$ [\text{truth telling constraint}] $$ $$ [\text{lender participation constraint}] $$
\\({\color[rgb]{0.7372549019607844,0.5882352941176471, 0.9019607843137255}\hat{v}^e}\\) is concave. Entrepreneurs are risk averse, but behave as risk neutral when financial markets are perfect.
--- class: left, middle ### The entrepreneur's intratemporal problem
$$ R(\theta,s),b^e,h^e =\arg \max \mathbb{E} \left\lbrace \hat{v}^e(R(\theta,s)q^e) | \theta\right\rbrace $$ subject to $$ R(\theta,s)q^e \leq f(\theta,k,h) - r^b(\theta)b^e - wh^e\qquad \forall \theta$$ in addition to $$ [\text{capital budget constraint}] $$ $$ {\color[rgb]{0.7372549019607844,0.5882352941176471, 0.9019607843137255}[\text{truth telling constraint}]} $$ $$ [\text{lender participation constraint}] $$
Quantitative model:
Imperfect state verification (Duncan and Nolan, 2019) Theory results can accommodate
Grossman Hart (1983), Krasa and Villamil (2000), Lacker and Weinberg (1989).
--- class: left, middle ### The entrepreneur's intratemporal problem
$$ R(\theta,s),b^e,h^e =\arg \max \mathbb{E} \left\lbrace \hat{v}^e(R(\theta,s)q^e) | \theta\right\rbrace $$ subject to $$ R(\theta,s)q^e \leq f(\theta,k,h) - {\color[rgb]{0.7372549019607844,0.5882352941176471, 0.9019607843137255}r^b(\theta)b^e} - wh^e\qquad \forall \theta$$ in addition to $$ [\text{capital budget constraint}] $$ $$ [\text{truth telling constraint}] $$ $$ [\text{lender participation constraint}] $$
\\({\color[rgb]{0.7372549019607844,0.5882352941176471, 0.9019607843137255}r^b(\theta)b^e}\\) is the net interest paid on loans. This is state-contingent, negative in default. Partially insures fluctuations in \\(\theta\\).
--- class: left, middle ### The entrepreneur's intratemporal problem
$$ R(\theta,s),b^e,h^e =\arg \max \mathbb{E} \left\lbrace \hat{v}^e(R(\theta,s)q^e) | \theta\right\rbrace $$ subject to $$ R(\theta,s)q^e \leq f(\theta,k,h) - r^b(\theta)b^e - {\color[rgb]{0.7372549019607844,0.5882352941176471, 0.9019607843137255}wh^e}\qquad \forall \theta$$ in addition to $$ [\text{capital budget constraint}] $$ $$ [\text{truth telling constraint}] $$ $$ [\text{lender participation constraint}] $$
\\({\color[rgb]{0.7372549019607844,0.5882352941176471, 0.9019607843137255}wh^e}\\) is the total wage bill. Workers are hired before the realisation of \\(\theta\\). Wages are not contingent on \\(\theta\\).
--- class: left, middle ### The entrepreneur's intertemporal problem
$$ v^e(q^e) = \max_{x^e,c^e,{q^e}'} \mathbb{E} \left\lbrace u^e(c^e) + \beta^e v^e({q^e}')\right\rbrace $$ subject to $$ {q^e}' = R(\theta,s)q^e - c^e - \int_{s'\in S} p(s') x^e(s') ds + {x^e}(s')$$
--- class: left, middle ### The entrepreneur's intertemporal problem
$$ v^e(q^e) = \max_{x^e,c^e,{q^e}'} \mathbb{E} \left\lbrace u^e(c^e) + \beta^e v^e({q^e}')\right\rbrace $$ subject to $$ {q^e}' = {\color[rgb]{0.7372549019607844,0.5882352941176471, 0.9019607843137255}R(\theta,s)}q^e - c^e - \int_{s'\in S} p(s') x^e(s') ds + {x^e}(s')$$
\\(\color[rgb]{0.7372549019607844,0.5882352941176471, 0.9019607843137255}R(\theta,s)\\) is an indirect gross return function. It is the result of privately optimal borrowing and hiring, from the entrepreneur's intratemporal problem.
--- class: left, middle ### The entrepreneur's intertemporal problem
$$ v^e(q^e) = \max_{x^e,c^e,{q^e}'} \mathbb{E} \left\lbrace u^e(c^e) + \beta^e v^e({q^e}')\right\rbrace $$ subject to $$ {q^e}' = R(\theta,s)q^e - c^e - {\color[rgb]{0.7372549019607844,0.5882352941176471, 0.9019607843137255}\int_{s'\in S} p(s') x^e(s') ds + {x^e}(s')}$$
\\({\color[rgb]{0.7372549019607844,0.5882352941176471, 0.9019607843137255}\int_{s'\in S} p(s') x^e(s') ds + {x^e}(s')}\\) captures trade in aggregate state contingent securities. Markets for aggregate risks are complete.
--- class: left, middle ### The entrepreneur's intertemporal problem
$$ v^e(q^e) = \max_{x^e,c^e,{q^e}'} \mathbb{E} \left\lbrace u^e(c^e) + \beta^e v^e({q^e}')\right\rbrace $$ subject to $$ {q^e}' = R(\theta,s)q^e - c^e - \int_{s'\in S} p(s') x^e(s') ds + {x^e}(s')$$
**Assumption (anonymity)** Entrepreneurs are anonymous across financial markets.
**Assumption (interior borrowing)** Intratemporal financial allocations \\(x^e(s')\\) are unconstrained.
**Assumption (regularity)** \\(R\\) invertible, continuously differentiable, convex.
--- class: left, middle ### The household's problem
$$ v(q) = \max_{x,c,h,{q}'} \mathbb{E} \left\lbrace u(c,h)+ \beta v({q}')\right\rbrace $$ subject to $$ q' = (1+r)q + wh - c - {\color[rgb]{0.7372549019607844,0.5882352941176471, 0.9019607843137255}\int_{s'\in S} p(s') x(s') ds + x(s')}$$
\\({\color[rgb]{0.7372549019607844,0.5882352941176471, 0.9019607843137255}\int_{s'\in S} p(s') x(s') ds + {x}(s')}\\) captures trade in aggregate state contingent securities. Markets for aggregate risks are complete.
--- class: left, middle ### Factor markets
$$\color[rgb]{0.7372549019607844,0.5882352941176471, 0.9019607843137255} l = \dfrac{\mathbb{E}_\Theta f(\theta,k,h)}{(1+r)q^e} $$ $$\dfrac{\mathbb{E}_\Theta [R(\theta,s)]}{1+r} = 1+l\tau$$ $$w = \mathbb{E}_\Theta f_h(\theta,k,h)( 1-\tau )$$
\\(\color[rgb]{0.7372549019607844,0.5882352941176471, 0.9019607843137255}\text{Leverage, }l\\)
--- class: left, middle ### Factor markets
$$ l = \dfrac{\mathbb{E}_\Theta f(\theta,k,h)}{(1+r)q^e} $$ $$\color[rgb]{0.7372549019607844,0.5882352941176471, 0.9019607843137255}\dfrac{\mathbb{E}_\Theta [R(\theta,s)]}{1+r} = 1+l\tau$$ $$w = \mathbb{E}_\Theta f_h(\theta,k,h)( 1-\tau )$$
\\(\color[rgb]{0.7372549019607844,0.5882352941176471, 0.9019607843137255}\text{The equity risk premium, }\\) \\(\color[rgb]{0.7372549019607844,0.5882352941176471, 0.9019607843137255}\rho := \dfrac{\mathbb{E}_\Theta [R(\theta,s)]}{1+r}\\)
--- class: left, middle ### Factor markets
$$ l = \dfrac{\mathbb{E}_\Theta f(\theta,k,h)}{(1+r)q^e} $$ $$\dfrac{\mathbb{E}_\Theta [R(\theta,s)]}{1+r} = 1+l\tau$$ $$\color[rgb]{0.7372549019607844,0.5882352941176471, 0.9019607843137255} w = \mathbb{E}_\Theta f_h(\theta,k,h)( 1-\tau )$$
\\(\color[rgb]{0.7372549019607844,0.5882352941176471, 0.9019607843137255}\text{Wages, }w\\)
--- class: left, top ### The competitive allocation of aggregate risk
$$ \dfrac{\beta^e\ {\mathbb{E}'_{\Theta}u^{e}}'({c^e}'(\theta'))}{{u^e}'(c^e)} = \dfrac{\beta\ u'({c}',h')}{u'(c,h)} $$
--- class: left, top ### Optimal macroprudential policy
$$ (1+\omega) \dfrac{\beta^e\ {\mathbb{E}'_{\Theta}u^{e}}'({c^e}'(\theta'))}{{u^e}'(c^e)} = \dfrac{\beta\ u'({c}',h')}{u'(c,h)} $$
Under optimal policy $$ \dfrac{\partial\omega}{\partial l},\dfrac{\partial\omega}{\partial \sigma} > 0.$$ Optimal macroprudential policy leans against - fluctuations in leverage, and - entrepreneurs' exposure to risk shocks. --- class: left, top ### The macroprudential externality
- Cyclical risk is a complement to downturn moral hazard. - Entrepreneurs accept too much cyclical risk, - amplifying the cost of moral hazard in downturns, - Arnott-Stiglitz: Regulate cyclical risk (macroprudential) --- class: left, middle # Optimal wage subsidy policy --- class: left, top ### Optimal wage subsidy policy - example with log utility
**Proposition** Let $$u(c,h) = \log c - \dfrac{h^{1+\psi}}{1+\psi},\qquad u^e(c^e) = \log c^e. $$ Optimal wage subsidy: $$\varsigma^* = \dfrac{\tau}{1-\tau} - {\color[rgb]{0.7372549019607844,0.5882352941176471, 0.9019607843137255}\hat{\lambda'}}(1-\beta^e) \dfrac{ 1+l\tau }{l(1-\tau)}$$ where $$\lambda = \frac{{u^e}'(\bar{c}^e)}{u_c(c,h)},\qquad\hat{\lambda}' := \frac{\lambda'-\lambda_0}{ \lambda'}.$$
--- class: left, top ### Optimal wage subsidy policy - example with log utility
Optimal wage subsidy $$\varsigma^* = {\color[rgb]{0.7372549019607844,0.5882352941176471, 0.9019607843137255}\dfrac{\tau}{1-\tau}} - \hat{\lambda'}(1-\beta^e) \dfrac{ 1+l\tau }{l(1-\tau)}$$
Optimal wage subsidy - completely offsets the financial wedge on impact,
--- class: left, top ### Optimal wage subsidy policy - example with log utility
Optimal wage subsidy $$\varsigma^* = \dfrac{\tau}{1-\tau} - {\color[rgb]{0.7372549019607844,0.5882352941176471, 0.9019607843137255}\hat{\lambda'}}(1-\beta^e) \dfrac{ 1+l\tau }{l(1-\tau)}$$ where $$\hat{\lambda}' := \frac{\lambda'-\lambda_0}{ \lambda'},$$ \\(\lambda' = \lambda\dfrac{\beta^e}{\beta}(1+l\tau).\\) \\(\hat{\lambda}'\\) is a summary statistic for history of past financial wedges.
Optimal wage subsidy - completely offsets the financial wedge on impact, - is moderated by past financial wedges.
--- class: left, top ### How the intervention works
*Benefit* Wage subsidies - complement firms' wealth, - encourage precaution during expansions, and - decreases financial frictions in downturns. - First order welfare gain.
*Cost* Wage subsidies - introduces a distortion between labour supply and demand, - Second order welfare cost. --- class: left, middle # Quantitative exercise --- class: left, top
### The entrepreneur combines their own wealth with borrowed wealth and labour. Contracts are endogenously incomplete. Entrepreneurs can hide income from external creditors. External creditors can audit the firm and uncover hidden income, but these audits are noisy. (Duncan and Nolan, 2019)
--- class: left, middle ### Exercise
- We estimate the model on US business cycle data, with no macroprudential or wage subsidy policy. - We add a wage subsidy, via a simple rule, and find the optimal simple rule and the associated welfare gain.
--- class: left, top ### Wage subsidy simple rule
We propose the following simple rule: $$ \varsigma = -\phi_\varsigma (y-y_0)$$ where \\(\varsigma\\) is the wage subsidy (tax if negative), and \\(y_0\\) is deterministic steady state output.
--- class: center, top #### Expected welfare effects of wage subsidy simple rules
Welfare gain is expressed as a share of business cycle welfare losses.
Shaded area indicates 90% credible interval. --- class: center, top #### Persistence of TFP shocks
Welfare gain is expressed as a share of business cycle welfare losses. --- class: left, middle ## Dixit's critique If the margin that the policy is acting on is distorted,
then the Arnott-Stiglitz logic doesn't necessarily hold. The cost of the tax distortion could be first order. We add 1. A static labour tax 2. Dynamic New Keynesian markups --- class: center, top #### With 40% constant labour tax
Welfare gain is expressed as a share of business cycle welfare losses.
Shaded area indicates 90% credible interval. --- class: left, top ### Interactions with monetary policy We derive a small-scale log-linear New Keynesian version of our model. The terms in black are as in Gali (2007). IS Curve $$({\color{blue}\zeta} + \gamma{\color{blue}-1})y = ({\color{blue}\zeta} + \gamma{\color{blue}-1})\mathbb{E}[y'] - {\color{blue}\zeta}(i - \mathbb{E}[\pi']) - {\color{blue}(\zeta-1)\varphi l - \gamma\Delta \varphi (\rho_\sigma-\varphi) \sigma}$$ Phillips Curve $$\pi = \beta \mathbb{E}[\pi'] + \lambda\left(\chi + \gamma - 1\right) y -\lambda\chi z {\color{blue}- \lambda \varsigma + \lambda(\zeta + \delta -1) l + \lambda(\delta_\sigma-\gamma\Delta\varphi)\sigma}$$ Leverage updating $${\color{blue}\zeta l = \left(\zeta-\varphi\right) L l + \gamma\Delta\varphi \sigma - (1+\gamma\Delta)\varphi L\sigma - (\gamma-1)(y-Ly)}$$ The policy variables are \\(i\\) and \\({\color{blue}\varsigma}\\). --- class: center, top #### Optimal policy
--- class: left, top ### Interactions with monetary policy
Proposition 2
Under joint optimal monetary and wage subsidy policy, the optimal path of inflation is zero in all periods \\(\pi_t = 0 \: \forall t\\).
Proposition 3
Let \\(\gamma>1\\). When the convexity of monitoring costs is relatively high (low), optimal output growth is lower (higher) when leverage is high, all else equal.
--- class: left, top ### Interactions with monetary policy
Proposition 2
Under joint optimal monetary and wage subsidy policy, the optimal path of inflation is zero in all periods \\(\pi_t = 0 \: \forall t\\).
- Would get the same result from a standard NK model with markup shocks and the same policy instruments.
Proposition 3
Let \\(\gamma>1\\). When the convexity of monitoring costs is relatively high (low), the optimal output growth is lower (higher) when leverage is high, all else equal.
- The convexity of monitoring costs generates a wedge between the social and private marginal costs of monitoring (see Hillier and Worral 1994 EJ). When high, the planner seeks a smoother path of leverage.
--- class: left, top ### Covid-19 wage subsidies
United Kingdom - Furlough scheme. - Not conditioned on firm outcomes, complements inside wealth. - Furloughed workers not permitted to work. - Not great stimulus. - Not much value for employers (not very prudential). - Seems to not have been enforced (maybe a good thing).
--- class: left, top ### Summary
We present a model where moral hazard generates a macroprudential externality. In lieu of aggregate demand externalities, there is still a role for fiscal stimulus. If the stimulus programme complements inside wealth, like a labour subsidy, then it will - encourage firms' prudence during the preceding expansion, and - reduce the costs of the moral hazard friction, - increasing welfare.