Maths in finance

[greedy banker]

If the trickle flows strongly enough, of course it is.

 

[Cid]

Of course it's theft, whether or not it is justified is another debate.

As for this 1% thingy - don't forget 0.01% of humans have probably contributed 99.99% to where we are today.

Regressive taxation is a massive problem though, I'll give you that.

Depends how you look at things... financial contributions are obviously essential, but somewhat pointless if you don't have the people who come up with ideas or a vast, skilled workforce...

 

[ymu]

As for this 1% thingy - don't forget 0.01% of humans have probably contributed 99.99% to where we are today.

That would require the productive 0.01% to have put in 10,000 times more hours than the unproductive 99.9%, which seems a tad unlikely.

If you mean someone like, say, Einstein contributed massive amounts to the sum of human knowledge and technological progress, I'd have to agree. But the people who grew his food, or made his clothes, or typeset his papers or did whatever needed to be done to enable the guy to devote his time to the frontiers of science, contributed just as much. They had a less glamorous and uncredited role, but that doesn't make the value of their contribution any less.

Not sure Einstein ever was particularly highly paid though. In fact I'm not sure there's much crossover between your 0.01% of contributors and the rich-list at all. Did you have any particular individuals in mind?

 

[ymu]

If the trickle flows strongly enough, of course it is.

Even the original proponents of trickle-down have admitted it was all bollocks. Anyone who supports the notion nowadays exposes themselves as a gullible fool. Sorry.

 

[Knotted]

Compare this with modern science which, in its laborious, often excruciating way, insists on identifying the mechanisms behind the empirical data. The empirical data are the starting point of research rather than the end point, because what is sought fundamentally is understanding.

I don't think that's true with respect to modern theoretical physics. Some of the models are too complicated to fully understand. Some of them are not even known to exist as mathematical objects.

 

[Knotted]

I think you're a bit naive about science tbh. Statistics is a key tool for applied scientists but very few of them have more than a superficial understanding of what they're doing. Psychologists are notorious for using ANOVA for *everything*; doctors are notorious for making it up as they go along; surgeons are notorious for not seeing the need to do any empirical research at all.

In my view statistics should be taught as part of every science degree. It shocks me that it isn't.

 

[kabbes]

I'm clearly too late into this thread, since it has already devolved into a slanging match. However, lbj, I'll attempt to answer your original question.

The statistical models I build are built from the ground up to attempt to simulate reality. Each element tends to be relatively straightforward. For example, I want to simulate future losses, so I might pick a Poisson distribution to represent claim numbers and then a Pareto distribution to represent claim amounts.

These elements then interact with each other. Each interaction is again built from the ground up. For example, I may have a reinsurance that kicks in at £50,000. So if a simulated loss is greater than £50,000, the reinsurance recovers the excess.

In this way the model is built up and built up. Each bit of it is well understood by me, because I have designed it to replicate the actual underlying process. You can argue about the parameters used and you can argue whether that specific distribution or methodology is appropriate, but this doesn't change the underlying logic.

So finally I end up with my all-singing, all-dancing stochastic simulation model. It will simulate tens of thousands of possible outcomes based on a myriad of statistical distributions for the input parameters. It will have all kinds of dependencies, from straightforward correlations to crazy copula tail dependencies. Each bit of it is easy to understand (so long as you have the right background, of course). I understand the model, I can explain it to others, I can justify it piece by piece.

But the answer -- ah, that's a different story. There would be no point if having the model if the answer were easy to predict. Essentially, the system is partially chaotic (hopefully it has sufficient inbuilt dampening to prevent actual chaos, but the concept is always there in the background). You build the model precisely to see what will happen if you change things. It's a bit like sticking a model aeroplane into a wind tunnel. I want to investigate the turbulence it creates. Essentially, I'm a statistical engineer at this point.

 

[ymu]

Nice description kabbes. It's exactly the same with any form of decision-analytic modelling. Physics deals with fixed underlying natural laws - much of science does not. Half of the effort expended on NIHCE appraisals, for example, is in working out why the industry model makes the treatment sooooo much cheaper than the academic one. It's usually easy to spot once you've unravelled the model - they've assigned a "fate worse than death" to the untreated state, or cherry-picked/manipulated the data to make the treatment seem more effective, or structured the model to double-count (or more) the benefits, or omit some of the costs.

Someone quoted a quant somewhere on one of these threads as saying that the CEOs weren't interested in models that told them they would lose - so obviously the quants have an incentive to produce ones that said they would win. Exactly the same as the scientists-for-hire that produce the industry models for NIHCE. They don't get it wrong because they don't understand the model - they get it wrong because they'll get paid for it.

Which is what is so utterly moronic about the crisis - the bankers were paying bonuses to people who had very different interests, all of which worked against the banks. Sell more mortgages = bonuses earned by selling them to sub-prime customers. Repackage sub-prime mortgages = bonuses for rating them much higher than they deserve. Sell packaged bad debt = bonuses for cleaning up the books. Buy packaged bad debt - bonuses for an apparently solid bottom line. All the while, creating a bubble in the exact same asset that is securing the dodgy debt in the first place.

 

[ymu]

In my view statistics should be taught as part of every science degree. It shocks me that it isn't.

I agree, but unfortunately it isn't really the answer. We used to teach medics no stats at all and they made it up as they went along. Now we teach them enough to critically appraise the literature and they think they can go off and do research without consulting anyone else - which is partly because that's exactly what we make them do with their projects. @ us

It takes several years to become professionally qualified as a statistician, and then only within the specialist area that you work in. There's no point demanding that of all applied scientists - they need to become good at other stuff. The key is multi-disciplinary teams - but it's quite a difficult structure to work out. It's not great being a statistician in a department full of doctors because you don't get to interact with many other statisticians. And you need to be stubborn as fuck to get the science taken seriously - they're often more interested in tailoring the proposal to a plausible amount of funding rather than the correct research design, or using the results that they like the look of, rather than the ones that actually answer the question.

 

[kabbes]

Indeed, ymu (responding to post above last one). And it is made all the more murky by the fact that the *more* you understand the model, the more you understand just how much potential model error is involved, which means that less strictly you follow what it is telling you, which means the more susceptible you are to juuuuust tweaking it a bit here and there to get the result you wanted in the first place...

 

[ymu]

Oh yes. We use probabilistic sensitivity analysis when possible to try to estimate the error better, but some of the models which most accurately represent the clinical situation are too complex to run with currently available computing power/algorithms.

 

[Kanda]

Real solutions are not forthcoming. How about moving away from a shareholding economy towards a mutual/partnership economy based on the success of building societies and companies like John Lewis.

Something that redistributes all profits to employees etc?

 

[littlebabyjesus]

@ Ymu and kabbes: Thanks for your answers. Interesting reading.

@ Kanda: Essentially, yes. To me shareholders are essentially leeches making money out of owning rather than producing. Issuing shares is of course one way of raising capital, but there must surely be other, less exploitative, ways to do that. The success of those cooperative models that do exist shows how it is possible to be successful without shareholders.

 

[Paulie Tandoori]

I'm clearly too late into this thread, since it has already devolved into a slanging match. However, lbj, I'll attempt to answer your original question.

The statistical models I build are built from the ground up to attempt to simulate reality. Each element tends to be relatively straightforward. For example, I want to simulate future losses, so I might pick a Poisson distribution to represent claim numbers and then a Pareto distribution to represent claim amounts.

These elements then interact with each other. Each interaction is again built from the ground up. For example, I may have a reinsurance that kicks in at £50,000. So if a simulated loss is greater than £50,000, the reinsurance recovers the excess.

In this way the model is built up and built up. Each bit of it is well understood by me, because I have designed it to replicate the actual underlying process. You can argue about the parameters used and you can argue whether that specific distribution or methodology is appropriate, but this doesn't change the underlying logic.

So finally I end up with my all-singing, all-dancing stochastic simulation model. It will simulate tens of thousands of possible outcomes based on a myriad of statistical distributions for the input parameters. It will have all kinds of dependencies, from straightforward correlations to crazy copula tail dependencies. Each bit of it is easy to understand (so long as you have the right background, of course). I understand the model, I can explain it to others, I can justify it piece by piece.

But the answer -- ah, that's a different story. There would be no point if having the model if the answer were easy to predict. Essentially, the system is partially chaotic (hopefully it has sufficient inbuilt dampening to prevent actual chaos, but the concept is always there in the background). You build the model precisely to see what will happen if you change things. It's a bit like sticking a model aeroplane into a wind tunnel. I want to investigate the turbulence it creates. Essentially, I'm a statistical engineer at this point.

And then a black swan come along and BOOM!!!! it blows the whole damned game out of the water

 

[Kanda]

@ Kanda: Essentially, yes. To me shareholders are essentially leeches making money out of owning rather than producing. Issuing shares is of course one way of raising capital, but there must surely be other, less exploitative, ways to do that. The success of those cooperative models that do exist shows how it is possible to be successful without shareholders.

I work for a company that redistributes all profits to employees, it's a Hedge Fund though

(Albeit one that only trades in stocks you or I could buy. Not any of these debt products etc, hence we have been pretty much untouched by this crunch)

 

[kabbes]

And then a black swan come along and BOOM!!!! it blows the whole damned game out of the water

It does if you have assumed that they can't exist, yes indeed. Many examples of just such a thing.

 

[Jonti]

I've seen black swans on the Thames down at Rotherhithe. Apparently they're not such an uncommon sight there.