Category Archives: Risk management

30/4/20: No, Healthcare Systems are Not Lean Startups, Mr. Musk

A tweet from @elonmusk yesterday has prompted a brief response from myself:

For two reasons, as follows, it is worth elaborating on my argument a little more:

  1. I have seen similar sentiment toward authorities' over-providing healthcare system capacity in other countries as well, including, for example in Ireland, where the public has raised some concerns with the State contracting private hospitals for surplus capacity; and
  2. Quite a few people have engaged with my response to Musk.
So here are some more thoughts on the subject:

'Lean startups' is an idea that goes hand-in-hand with the notion that a startup needs some organic growth runway. In other words, it needs to ‘nail’ parts of its business model first, before ‘scaling’ the model up. ‘Nailing’ bit is done using highly scarce resources pre-extensive funding (which is a ‘scaling’ phase). It makes perfect sense for a start up, imo, for a startup.

But in the ‘nailing’ stage, when financial resources are scarce, the startup enterprise has another resource is relies upon to execute on a ‘lean’ strategy: time. Why? Because a ‘lean’ startup is a smaller undertaking than a scaling startup. As a result, failure at that stage carries lower costs. In other words, you can be ‘lean’ because you are allowed to fail, because if you do fail in that stage of development, you can re-group and re-launch. You can afford to be reactive to news flows and changes in your environment, which means you do not need to over-provide resources in being predictive or pro-active. Your startup can survive on lean funding.

As you scale startup, you accumulate resources (investment and retained earnings) forward. In other words, you are securing your organization by over-providing capacity. Why? Because failure is more expensive for a scaling startup than for a 'lean' early stage startup. The notion of retained and untilized cash is no longer the idea of waste, but, rather a prudential cushion. Tesla, Mr. Musk's company, carries cash reserves and lines of credit that it is NOT using at the moment in time precisely because not doing so risks smaller shocks to the company immediately escalating into existential shocks. And a failure of Tesla has larger impact than a failure of small 'lean' startup. In other words, Mr. Musk does not run a 'lean startup' for a good reason. Now, in a public health emergency with rapid rates of evolution and high degree of forecast uncertainty, you cannot be reactive. You must allocate resources to be pro-active, or anticipatory. In doing so, you do not have a choice, but to over-supply resources. You cannot be ‘lean’, because the potential (and highly probable) impact of any resource under-provision is a public health threat spinning out of control into a public health emergency and a systemic shock. ‘Lean’ startup methods work, when you are dealing with risk and uncertainty in a de-coupled systems with a limited degree of complexity involved and the range of shocks impact limited by the size of the organization/system being shocked. Public health emergence are the exact opposite of such a environment: we are dealing with severe uncertainty (as opposed to risk) with hugely substantial impacts of these shocks (think thousands of lives here, vs few million dollars in investment in an early stage start up failure). We are also dealing with severe extent of complexity. High speed of evolution of threats and shocks, uncertain and potentially ambiguous pathways for shocks propagation, and highly complex shock contagion pathways that go beyond the already hard-to-model disease contagion pathways. So a proper response to a pandemic, like the one we are witnessing today, is to use an extremely precautionary principle in providing resources and imposing controls. This means: (1) over-providing resources before they become needed (which, by definition, means having excess capacity ex-post shock realization); (2) over-imposing controls to create breaks on shock contagion (which, by definition, means doing too-much-tightening in social and economic environment), (3) doing (1) and (2) earlier in the threat evolution process rather than later (which means overpaying severely for spare capacity and controls, including - by design - at the time when these costs may appear irrational). And (4), relying on the worst-case-scenario parameterization of adverse impact in your probabilistic and forecasting analysis and planning. This basis for a public health threat means that responses to public health threat are the exact opposite to a ‘lean’ start up environment. In fact they are not comparable to the ‘scaling up’ start up environment either. A system that has a huge surplus capacity left in it, not utilized, in a case of a start up is equivalent to waste. Such system’s leadership should be penalized. A system that has a huge surplus capacity left un-utilized, in a case of a pandemic is equivalent to the best possible practice in prudential management of the public health threat. Such system’s leadership should be applauded.

And even more so in the case of COVID pandemic. Mr. Musk implies something being wrong with California secured hospital beds capacity running at more than double the rate of COVID patients arrivals. That's the great news, folks. COVID pandemic carries infection detection rates that double the population of infected individuals every 3-30 days, depending on the stage of contagion evolution. Earlier on, doubling times are closer to 3 days, later on, they are closer to 30 days. But, utilization of hospital beds follows an even more complex dynamic, because in addition to the arrival rates of new patients, you also need to account for the duration of hospital stay for patients arriving at different times in the pandemic. Let's be generous to sceptics, like Mr. Musk, and assume that duration-of-stay adjusted arrivals of new patients into the hospitals has a doubling time of the mid-point of 3-30 days or, close to two weeks. If California Government did NOT secure massively excessive capacity for COVID patients in advance of their arrival, the system would not have been able to add new capacity amidst the pandemic on time to match the doubling of new cases arrivals. This would have meant that some patients would be able to access beds only later in the disease progression period, arriving to hospital beds later in time, with more severe impact from the disease and in the need of longer stays and more aggressive interventions. The result would have been even faster doubling rate in the demand for hospital beds with a lag of few days. You can see how the system shortages would escalate out of control.

Running tight supply chains in a pandemic is the exact opposite to what has to be done. Running supply capacity at more than double the rate of realized demand is exactly what needs to be done. We do not cut corners on basic safety equipment. Boeing did, with 737-Max, and we know where they should be because of this. We most certainly should not treat public health pandemic as the basis for cutting surplus safety capacity in the system.

26/6/16: Black Swan ain’t Brexit… but

There is a lot of froth in the media opinionating on Brexit vote. And there is a lot of nonsense.

One clearly cannot deal with all of it, so I am going to occasionally dip into the topic with some comments. These are not systemic in any way.

Let's take the myth of Brexit being a 'Black Swan'. This goes along the lines: lack of UK and European leaders' preparedness to the Brexit referendum outcome can be explained by the nature of the outcome being a 'Black Swan' event.

The theory of 'Black Swan' events was introduced by Nassin Taleb in his book “Black Swan
Theory”. There are three defining characteristics of such an event:

  1. The event can be explained ex post its occurrence as either predictable or expected;
  2. The event has an extremely large impact (cost or benefit); and
  3. The event (ex ante its occurrence) is unexpected or not probable.

Let's take a look at the Brexit vote in terms of the above three characteristics.

Analysis post-event shows that Brexit does indeed conform with point 1, but only partially. There is a lot of noise around various explanations for the vote being advanced, with analysis reaching across the following major arguments:

  • 'Dumb' or 'poor' or 'uneducated' or 'older' people voted for Brexit
  • People were swayed to vote for Brexit by manipulative populists (which is an iteration of the first bullet point)
  • People wanted to punish elites for (insert any reason here)
  • Protests vote (same as bullet point above)
  • People voted to 'regain their country from EU' 
  • Brits never liked being in the EU, and so on
The multiplicity of often overlapping reasons for Brexit vote outcome does imply significant complexity of causes and roots for voters preferences, but, in general, 'easy' explanations are being advanced in the wake of the vote. They are neither correct, nor wrong, which means that point 1 is neither violated nor confirmed: loads of explanations being given ex post, loads of predictions were issued ex ante.

The Brexit event is likely to have a significant impact. Short term impact is likely to be extremely large, albeit medium and longer term impacts are likely to be more modest. The reasons for this (not an exhaustive list) include: 
  • Likely overshooting in risk valuations in the short run;
  • Increased uncertainty in the short run that will be ameliorated by subsequent policy choices, actions and information flows; 
  • Starting of resolution process with the EU which is likely to be associated with more intransigence vis-a-vis the UK on the EU behalf at the start, gradually converging to more pragmatic and cooperative solutions over time (what we call moving along expectations curve); 
  • Pre-vote pricing in the markets that resulted in a rather significant over-pricing of the probability of 'Remain' vote, warranting a large correction to the downside post the vote (irrespective of which way the vote would have gone); 
  • Post-vote vacillations and debates in the UK as to the legal outrun of the vote; and 
  • The nature of the EU institutions and their extent in determining economic and social outcomes (the degree of integration that requires unwinding in the case of the Brexit)
These expected impacts were visible pre-vote and, in fact, have been severely overhyped in media and official analysis. Remember all the warnings of economic, social and political armageddon that the Leave vote was expected to generate. These were voiced in a number of speeches, articles, advertorials and campaigns by the Bremainers. 

So, per second point, the event was ex ante expected to generate huge impacts and these potential impacts were flagged well in advance of the vote.

The third ingredient for making of a 'Black Swan' is unpredictable (or low predictability) nature of the event. Here, the entire thesis of Brexit as a 'Black Swan' collapses. 

Let me start with an illustration: about 18 hours before the results were announced, I repeated my view (proven to be erroneous in the end) that 'Remain' will shade the vote by roughly 52% to 48%. As far as I am aware, no analyst or media outfit or /predictions market' (aka betting shop) put probability of 'Leave' at less than 30 percent. 

Now, 30 percent is not unpredictable / unexpected outcome. It is, instead, an unlikely, but possible, event. 

Let's do a mental exercise: you are offered by your stock broker an investment product that risks losing 30% of our pension money (say EUR100,000) with probability of 30%. Your expected loss is EUR9,000 is not a 'Black Swan' or an improbable high impact event, but instead a rather possible high impact event. Conditional (on loss materialising) impact here is, however, EUR30,000 loss. Now, consider a risk of losing 90% of your pension money with a probability of 10%. Your expected loss is the same, but low probability of a loss makes it a rather unexpected high impact event, as conditional impact of a loss here is EUR90,000 - three times the size of the conditional loss in the first case. 

The latter case is not Brexit, but is a Black Swan, the former case is Brexit-like and is not a Black Swan event. 

Besides the discussion of whether Brexit was a Black Swan event or not, however, the conditional loss (conditional on loss materialising) in the above examples shows that, however low the probability of a loss might be, once conditional loss becomes sizeable enough, the risk assessment and management of the event that can result in such a loss is required. In other words, whether or not Brexit was probable ex ante the vote (and it was quite probable), any risk management in preparation of the vote should have included full evaluation of responses to such a loss materialising. 

It is now painfully clear (see EU case here:, see Irish case here: that prudent risk management procedures were not followed by the EU and the Irish State. There is no serious contingency plan. No serious road map. No serious impact assessment. No serious readiness to deploy policy responses. No serious proposals for dealing with the vote outcome.

Even if Brexit vote was a Black Swan (although it was not), European institutions should have been prepared to face the aftermath of the vote. This is especially warranted, given the hysteria whipped up by the 'Remain' campaigners as to the potential fallouts from the 'Leave' vote prior to the referendum. In fact, the EU and national institutions should have been prepared even more so because of the severely disruptive nature of Black Swan events, not despite the event being (in their post-vote minds) a Black Swan.

15/1/16: Household Debt Sustainability in One Chart?

Here is a neat chart plotting household debt against long term interest rates in an attempt to visualise property prices in affordability / sustainability context:

Source: @resi_analyst

Irish progression is poor by debt measure, and is sustained (barely) by low interest rates, even post-deleveraging.