An interactive Total Covid Cost model
All models are wrong but some are useful
**An improved version of this model, modelling test and trace, and vaccine rollout, is available for download from here**
Summary
The second wave of Covid cases is gathering pace. And so is the debate about how policymakers should respond. One side asserts that lockdowns are needed while the other says the cure is worse that the disease and we should ease restrictions to save livelihoods.
Beneath this polarised debate lies a variety of assumptions about things like the economic consequences of the virus or the suppression measures that would control it, the infection fatality rate and value of a life. Part of the reason it’s hard to build consensus on the right policy response is that it’s impossible to have a proper debate without surfacing these assumptions and trade-offs.
What if we had a basic model that allowed everyone in the debate to put numbers on their assumptions and assertions, and quantify the impact of their preferred policy response?
This blog is about an attempt to do that with my Total Covid Cost model here (in Excel). Download it, make the inputs and assumptions your own, and run your own scenarios to figure out what the government should do. (Oh and let me know if you find any gremlins in the spreadsheet or think it should be put together differently!)
Beyond the trade off
All policy options facing the government right now are bad. These are horrific decisions for policymakers to have to make about how to minimise the terrible toll of the virus in lives and livelihoods. But someone has to make them. Preferably they do it with the best available evidence and, critically, the sound grasp of how the economy and the virus interact.
It has become common to hear people assert that there is a trade-off between ‘the economy’ and controlling the virus. But as Simon Wren Lewis, Tony Yates, Tim Harford, the IMF and others have explained, while restrictions can damage the economy so can the virus: if we allow it to spread and deaths to mount people will be fearful of going out. The result will most likely be both a horrific death toll and crushed economic activity anyway. The IMF’s recent analysis concluded that the majority of the reduction in mobility in advanced economies during the first three months of the pandemic was attributable to this ‘fear factor’. On this view, restrictions on activity now are in fact a way of saving the economy and preventing even more lost jobs and bankrupt businesses.
What’s the best policy?
To decide the optimal strategy in the face of a second wave, we need to think through the direct impact of any set of suppression measures on the economy, but also the impact that rising cases and deaths would have on the economy in their absence. We also need to take account of the health costs of any given strategy: the deaths that Covid will cause, its longer-term health impact for some survivors, and the non-Covid health costs that result from NHS capacity being taken up caring for Covid patients.
Among academics there is growing interest in developing combined epidemiological and macroeconomic models to derive policy conclusions. But the utility of such models for policymakers seems limited in the teeth of the crisis. First, they are ‘black box’ exercises as far as policymakers are concerned, requiring us largely to take the results on faith. Second, our lack of understanding about this particular virus makes it difficult to have any sense of whether model-generated results of, say, the impact of lockdown measures or death rates on economic activity are very accurate.
What would be more useful for policymakers — and for the wider debate — is a simpler model that captures the key feedback effects, and allows users to populate it with their own assumptions about the strength of those relationships. That way we could begin to understand what one would need to believe to think that easing restrictions, or imposing a hard national lockdown, is the best approach.
That’s that I’ve attempted to do with this spreadsheet. You can download it and play around with the inputs and assumptions to figure out what you think the policy stance should be.
How does the model work?
The basic idea is illustrated in the diagram below. More detail on the inputs is available in the notes and sources tab in the workbook.
In the absence of any policy measures, for the purposes of the model, the level of GDP determines the rate of spread of the virus, the ‘organic R’. Without any restrictions in place this organic R becomes the ‘prevailing R’ in society, which reflects how fast the virus is spreading and, after a lag, the number of deaths that result. The rate of daily fatalities determines how fearful people are, which feeds back to GDP: the more fearful people are the more GDP falls (you can change the sensitivity).
If policymakers decide to act, they pick a ‘policy R’ number and implement whatever measures are necessary to achieve it. If policy R is lower than organic R, then policy R dictates the rate of spread of the virus. Those restrictions also drive GDP lower by directly prohibiting some forms of economic activity. The spread of the virus is modelled using a simple SIR (Susceptible Infectious Recovered) model.
GDP costs are measured relative to the pre-pandemic normal, in one-day segments. Covid health costs are measured using valuations for a ‘quality-adjusted life year’ from HMT’s Green Book guidelines (change it if you want), using some assumptions about the average number of ‘quality-adjusted life years’ lost per death. Non-covid health costs are proxied by healthcare displaced from the NHS as covid cases rise.
All of the key assumptions are for users to decide. The baseline parameters that I’ve put in the calculation represent my best guess, based on various bits of evidence that we have — as far as possible I’ve provided reasoning and sources in the ‘notes and sources’ tab. Inevitably there’s a lot of guestimation here.
Scenarios
In this initial version of the model I have five scenarios based on what policymakers decide to do with ‘policy R’ (you can create others):
- S1: Tighten restrictions to achieve R=0.9 immediately
- S2: Ease restrictions such that policy R=3 — the fear factor then keeps the ‘organic’ R around 2
- S3: Tighten restrictions to R=0.9 with a reformed Test and Trace, from 1 December, halving the GDP impact of the measures required to achieve that
- S4: Delay tightening restrictions to R=0.9 until the start of November
- S5: Introduce a circuit-breaker that reduces R to 0.7 immediately for four weeks, with policy R reverting to 0.9 from 15 November and no improvement in test and trace.
Results
The first chart below shows total cost under each scenario, and the second breaks down the components. On the baseline assumptions, the total cost (covid health, non-covid health and economic costs) is about twice as large on a strategy of easing as it is under a suppression (but not lockdown) strategy. However, a 4-week circuit breaker comes out as the second most costly option.
Economic cost comes out highest for the circuit breaker scenario (partly because of the assumption of continued suppression measures after the lockdown period). Total costs are by far the largest under the easing strategy because of the loss of life due to covid and other non-covid health damage, but also because of impact the mounting death would have on economic activity.
The number of covid deaths varies significantly between scenarios, but the easing strategy is by far the worst. This is partly for the obvious reason that the virus spreads rapidly without policy measures to control it. But it’s also because the number of cases accelerates so rapidly that it overshoots the herd immunity threshold by a long way.
To give a sense of how the different elements of cost vary over time, we can compare the suppression (S1) and easing (S2) scenarios. Initially the economic cost of easing is lower than under suppression because the fear factor weighs on the economy less than suppression measures would. But as Covid spreads, deaths mount along with non-covid health costs as health services reach capacity. The economic cost of easing also rises as the growing number of daily deaths causes people to hunker down.
All sorts of other scenarios are possible of course, so I’ll blog further on some of the interesting things I’ve found.
Conclusions
As statistician George Box is supposed to have said, ‘all models are wrong but some are useful.’ Hopefully this one is wrong in a way that helps users cut through a polarised debate by clarifying their priors. None of the results presented here should be taken as firm conclusions. Instead they are initial findings from a basic model using best-guess inputs that reasonable people will differ on.
Changing key parameters like the fatality rate (for example if shielding works), the QALY value, the fear factor, or the assumed GDP loss for any given policy R can alter the scale and composition of the costs.
Please download it and have a play around with the spreadsheet! Even better, post comments on what you find. I’d love to hear whether you think the model structure seems right, whether there are any errors in it, and what conclusions you draw from making different assumptions.
With your help, hopefully the next iteration will give more accurate results that could help to cut through some of the assertions made on both sides of the ‘lockdown’ debate, and contribute to making better policy.
