A few weeks ago we were invited to a presentation at one of the major investment banking houses in Sydney .. all about the “outlook for equities” and what financial products might best suit investors needs.

Some of those products are quite complex – the progeny of financial engineering – with conditions and guarantees and triggers and critical events, and they present quite an interesting problem from the research perspective and the (investor) evaluation perspective : I’ll talk about those two streams a bit later on.

Firstly though, let me ponder on the presentation of the “outlook for equities”, given by a respected economist. I should confess that, despite my wife having an Economics degree and having myself done some work in Econometrics (towards a PhD thesis along the lines of individual and group level discrete choice models), I have never quite understood generic macroeconomic reasoning – that is, when it is model free..

I suppose I was expecting quantitative forecasting and some new and beautiful structural time series applications.

If you say “the future” to me I immediately think .. ok, let’s use the state of the art to make some projections and come up with a distribution of likely outcomes based on a rigorous analysis of all the data to hand. I guess I don’t quite grasp why one wouldn’t do that, given the many tools available, and an established set of principles and accepted practices – you can see my review of “Principles of Forecasting” in Financial Engineering News ( http://www.fenews.com/fen25/aitchison.html ) if you are interested.

If you are talking “outlook” that implies medium term and some of the simpler time-series methods become, arguably, less directly relevant, because of the way the prediction intervals fan out.

What we heard was mostly, however, “scenarios”.

A scenario is something along the lines of “IF Chinese manufacturing and export continues at its current rate, THEN we believe XYZ company should continue to do well”.

My first reaction to this is that it is insufficiently rigorous. It is in the form of a rule (albeit not one that had been discovered from data, as far as I could tell), so that is a good thing. Rule based approaches are good and are common in machine learning – well, the concept of conditionality is a good thing too, but we have no measure of the quality (success rate) of that rule.

Now, the next problem is that no probability is attached to that rule. To be fair – and indeed, I am not trying to be critical, just looking at what was presented to us – the statement was qualified to read “IF Chinese manufacturing and export continues at its current rate, AND we believe it will, THEN we believe XYZ company should continue to do well”.

Good!

A “degree of belief” has entered the arena.. Bayesians will be happy. But I would be happier if this degree of belief had some quantitative expression .. like an expected probability (of the condition being true) of 60%, lower decile at 40% .. I might then argue with that assessment, but at least I would have something to argue with/simulate against.

Also, I would have liked an ELSE clause in this rule. IF … THEN .. ELSE.

ELSE to capture the downside.

I had, for some reason, rather forgotten about Nassim Nicholas Taleb (NNT) and his “Fooled By Randomness” and “Black Swan” (rare events .. but note the recent contribution of Andrew Gelman on one of the sister blogs in the blogroll to estimating the probability of rare or never experienced events). But I recollected that he at one stage was heavily into puts (the right to sell) on a stock that he believed was more likely to go up than down, but if it did go down it would go down massively. This is the ELSE, the art and science of dealing with the unexpected.

The final problem .. a fuzzy and probably inappropriate output measure. Measuring, or making predictions about, the wrong thing is all too common and a trap into which we all fall. In this case it is obvious, in other cases not. What do I care if XYZ company “does well”?

I care not a fig.

I care about whether I can make money out of the prediction that I have just received. And since many people simultaneously received that same prediction then the profit seems less than certain – however, some savvy people at the presentation pointed out that not all recipients of “sure thing” information act upon it, and I suspect that is true. A related and possibly more serious dampener on the likely profitability of this implied trading rule, and it is an implied trading rule of the form “buy now because we think it is going to go up”, is the possibility that the market has already factored in the probable distribution of outcomes and that XYZ shares are “fully priced”. There is some heavy duty analysis that could be done on that, and some logical and semantic clarification required, but let’s not get into that. (It is related to valuing news, some other day perhaps).

OK, a bit of deconstruction is simple enough and somewhat useful. An investor could quite probably use the deconstructed version of this prediction and some simulation to adjust her personal probability distribution of the outcomes of an investment in XYZ and the corresponding appropriate portfolio weight. And the downside protection strategy.

Let me step back a bit.

My business is data analytics, the art and science of making the best you can from the data (hard and soft, knowledge and history and personal priors included) that you have at hand, or could be sensibly collected.

I wanted to start this weblog the way I mean to go on. That is, to examine the world of good and bad data and propositions and evidence that we are immersed in, to suggest ways of dealing with that – the present reality, how we might do enlightened and disciplined analyses of where we find ourselves and what we have – and ways of going forward.

Ways of analyzing existing data and of designing research studies to inform our intuition, to help us know what we don’t know, perhaps even to understand why we don’t know it.

In case that sounds overly precious, let me put it another way .. if we have some data, we can do things with it and in the process of doing those analyses we are dragged kicking and screaming to a realization of our ignorance. And if we don’t have any data – but almost always we do, of some form, my intuition your intuition and the synergy/conflict of those understandings is some data – then we can think about ways of getting good or at least better data, and we will be informed by the very process.

There never is an answer, other than the obvious which becomes obvious through the process. And from the now obvious, we move forward.

There are no outcomes other than increased understanding, and new ideas and new horizons.

Research – data analytics – is an enabler. It feeds you, your intellect, your work.

So, as I said, I wanted to start out as I meant to finish.

I am interested in the finance industry, as an exemplar of a complex world and the players in it , and from this small start I wanted to proceed to discuss how one might research consumer and advisor acceptance of complex products, perhaps those products engineered to match a given scenario.

Research and data analysis is always grounded in the now. We attended a seminar, it gave rise to a bunch of ideas spinning off the presented content and the suggested actions.

Next blog will, most probably, be about how one might – as a financial services provider – frame a research program and how we can meld pre-existing market data with the research outcomes to provide better estimates, how the competing alternatives can be visualized and targetted. I also plan to do some blogging on how an individual can assess complex financial products.