Thatcher was right. This is my reaction to reports that Dominic Cummings regards Brexit as a way of getting out of limits on state aid and so enabling the government to funnel cash towards the UK's own future big tech companies.

The problem with this is that, as Thatcher said, the state cannot pick winners. Corporate growth is largely unpredictable. "The stochastic element is predominant" concluded Alex Coad in one survey. "Corporate growth rates are random" found Paul Geroski in another (pdf). "There is low predictability [of corporate earnings growth] even with a wide variety of predictor variables" found (pdf) Louis Chan, Josef Lakonishok and Jason Karceski in a third.

This fact explains why fund managers don't beat the market. David Blake has concluded: "the vast majority of fund managers in our data set were not simply unlucky, they were genuinely unskilled." It's also why the performance of venture capital trusts has been very variable. In the last five years, one fund trebled investors' money but two others lost half of it. A big reason for this is that, as Charlie Cai has shown, it is prodigiously difficult for companies to grow and investors often under-estimate all the obstacles to growth.

Investors cannot, then, pick winners. Success is unpredictable. And not just in the corporate sector. It's true in films: United Artists and Universal both rejected Star Wars. It's true in publishing: J.K Rowling and C.S Lewis saw their manuscripts rejected many times. And in music: Elvis Presley was told to go back to truck-driving and Decca famously rejected the Beatles on the grounds that "guitar groups are on the way out."

As William Goldman said, nobody knows anything." In particular - as my examples show - extreme success seems unpredictable.

This isn't to say that Thatcher's laisser-faire approach to industrial policy was right. There's much that governments can do: fund and promote basic research; ensure that start-ups and small firms aren't held back by a lack of finance or by the wrong intellectual property regime; ensure a good supply of skilled workers, some from overseas; and so on.

What they cannot do, though, is spot winners.

This is because Thatcher's mentor Friedrich Hayek was right. Markets are complex processes, knowledge of which is dispersed, fragmented, tacit and not collectable by a single mind:

In the study of such complex phenomena as the market, which depend on the actions of many individuals, all the circumstances which will determine the outcome of a process...will hardly ever be fully known or measurable...[We] cannot acquire the full knowledge which would make mastery of the events possible.

The issue here is a fundamental one - what can governments know? - that should be behind many issues in economic policy. For example:

- How can we run counter-cyclical policy, given that nobody can reliably predict recessions? For me, this is an argument for stronger automatic stabilizers such as better welfare benefits. It's also a case for having the right policy infrastructure in place, so that (for example) the Bank of England can quickly write us all a cheque.

- Do local authorities really have sufficient knowledge of what worthwhile things need doing, and the ability to scale these up and down, to make a job guarantee feasible? For me it is this question, rather than macroeconomic considerations, that is key to the desirability of scheme.

- Can governments know enough about low-income families to target support to them? Many, of example, move in and out of work often and see their hours fluctuate. Can we design a benefit system that responds to such variation, or is a high citizens income a simpler alternative?

- How best should banks be regulated? The trend has been towards increased complexity. But Andy Haldane has argued:

As you do not fight fire with fire, you do not fight complexity with complexity. Because complexity generates uncertainty, not risk, it requires a regulatory response grounded in simplicity, not complexity.

It might be that high capital requirements, or even nationalization, are simpler alternatives.

- Can bosses know enough to control large organizations? If, on the other hand, knowledge of the organization is also fragmented, there's a case for greater workers' say.

Hayek's question - what can governments know? - should then be a fundamental, ubiquitous one.

But it's not. This, I fear, is due partly to a selection effect. Politics and opinion-mongering selects in favour of bumptious prats who over-estimate their knowledge. Which is an obstacle to good policy.

Cummings is, I think, wrong to think the state can pick winners. His error is not an idiosyncratic one, however, but rather the product of systemic dysfunction.