Predictive analytics, CEP and your local mechanic

Let the indignant comments begin, as a loonie states that “Predictive analytics is a capability based on complex event processing (CEP).” Post is here.

They say that one way to tell that the stock market is over-hyped is when the guys repairing cars down at the local garage are talking about it. The assumption being that your local mechanic doesn’t know the first thing about the stock market and is just quoting some stuff from the news.

Now recently, I’ve been reading blog posts all about predictive analytics coming from people who have clearly never used a predictive analytic in their lives. This is particularly true in the CEP space, where there is so much posting about predictive analytics and machine learning from people who don’t seem to know the first thing about these topics. The above post is a case in point.

So when people who don’t know much about the subject are talking so much about predictive analytics, making statements and assessing their strategic value – I wonder if the public is getting the wrong idea about this stuff. Remember that, to a large degree, over reliance on analytics caused the credit crisis.

Sure analytics are the future, but let’s not jump directly to so many applications. Before you dive into analytics, you need to set a solid foundation for ongoing critical analysis of the effectiveness of your analytics.

A little knowledge about analytics is a dangerous thing. Don’t assume that you can set up some analytics, let them run, and get good data forever. Think before you jump into analytics. And maybe also before you post about them as well.

Matlab, Mathematica in the cloud

I was just looking at Matlab and Mathematica on the Amazon cloud. Along with all the grid activity in the Amazon cloud, we can now farm out our parallel data processing tasks from analysis packages. I expect a guide to parallel R in the cloud soon. And as usual, SAS will follow along in another 5 years.

Fun use of Esper: Alerts4All

Alerts4All uses Esper and the AWS cloud to do some of those streaming calculations that you get from fancy quoting applications. So far, it’s just got the usual stuff, but it’s still cool.

You can get a free beta account through this link.

Seminar on Computational Finance with R at Columbia

See announcement below for an interesting seminar. Inconvenient time, though.

Computational Finance with R
http://www.stat.columbia.edu/pages/ComputationalFinance/index.html

Department of Statistics organizes a workshop about using statistical
computing with R in finance. The conference would like to bring
together both academics and practitioners, and it is open to public.
Admission is free, however we require that the participants register
in advance.

Registration Link
http://www.stat.columbia.edu/pages/ComputationalFinance/register.html

The conference is co-sponsored by REvolution Computing
(http://www.revolution-computing.com/home)

Schedule:

1:45 – 2:00PM   Refreshments

2:00 – 2:05PM   Opening Remarks

2:05 – 2:40PM   Whit Armstrong – Discount Curve Construction with fts,
RLIM, and RFincad (KLS Diversified Asset Management)

2:40 – 3:15PM   Anthony Brockwell – Quantitative Trading in Practice
(Horton Point LLC)

3:15 – 3:50PM   Bryan Lewis – High Performance R with Rpro (REvolution
Computing)

3:50 – 4:05PM   Coffee Break

4:05 – 4:40PM   Scott Payesur – Comparing Multivariate GARCH models
using Realized Covariance (UBS Asset Management)

4:40 – 5:15PM   Peter Carl and Brian Peterson – Performance Analysis
in R (PerformanceAnalytics)

5:15 – 5:50PM   Jeff Ryan – Quantmod Package (Quantmod)

6:00 – 6:30PM   Closing Reception

Analytics with dashboards: an iterative process

I just read this short post by James Taylor about analytics over dashboards.

Anyone who’s ever tried to fit a dozen models to a data set without improvement, only to have someone walk by the surface plot on their screen and instantly intuit a better model – knows the power of visualization. And if you do analytics or statistics, you have at least one story like this.

I’d bet money that as analytics are increasingly involved in running a business, the dashboard will become more integral, not less. But these will not be simple dashboards with stupid little dials showing how many customers are on the web site right now. Those dashboards are fast becoming a thing of the past because, as Mr. Taylor points out, who would watch that dial for any length of time?

You may not want to watch a dial with the number of website users, but how about a daily graph showing predicted visitors over the next month/quarter/year, side by side with a plot showing how accurate these predictions have been in the past 12 months? Now we’re talking.

In the next few years, we will see more and more dashboards that integrate tightly with analytics. They will distill information from analytics and they will incorporate some of the exploratory features that model builders use every day (visualization of residuals perhaps?) They will drill down into various components of the analytics and help visualize which methods are working, which aren’t and why.

Who will use dashboards that merge with analytics? The business users will use them to understand the performance and the output of their models.  The model builders will use them to intuit patterns.  And everyone will use them to fix problems in the analytics when predictions go wrong or optimization no longer produces the optimal result.

So I’m not quite ready to declare the death of the dashboard. But as analytics move more and more into the mainstream, dashboards certainly need to keep pace.

CEP software and competitive advantage

In another followup to a recent post by Marc (here) about some problems with the marketing of Complex Event Processing (CEP) software, I think that the kind of article that Marc rails against comes about for exactly one reason:

It’s better to sell a competitive advantage than a software license.

If a vendor can cast their product as a competitive advantage rather than simply a tool, it will be easier to justify a high price. They want you to buy the idea of real-time risk management, because this is something that just might impact your bottom line. It’s much less believable to claim that the purchase of just a software tool will materially affect your firm.

The vendors that started in the capital markets space mostly started by renting licenses. Meaning that you pay quarterly for the continued right to use the software. This works out very well for a firm using the software to trade. As long as the trading is making money, you pay for the software. They justify a nice and high price by claiming it as an expense of a profitable portion of your trading business. This is something that a trader understands: Your software lets me make more money, and I pay you a nice chunk of that money.

But what happens for sales that don’t involve trading? How can they justify such a high price for a project that doesn’t generate revenue? The answer is to show how the software will have some other strategic impact. This is pretty much the only way for vendors to keep competition from whittling prices down to levels that can’t support a decent profit.

So when we see strange claims by CEP vendors, what we are really seeing are trial balloons for arguments about competitive advantage. The vendors hope that if just one balloons floats, they can ride it to “strategic” sales to the business side.

CEP vendors need to drum up interest in real-time projects as a strategic or competitive advantage for the firm. Front office capital markets teams have been doing this for years. Competitive advantage is the argument supporting the relative autonomy of most front office IT organizations.  So with both the vendors and the buyers trying to show how strategic their projects are, you would think that the vendors would have an easier time of this process. But so far, they seem to be making somewhat of a mess of things every time they float one of these trial balloons (there are probably differing perspectives on that).

Now just how much of a competitive advantage can come from CEP software? Can competitive advantage come from building in analytics? Or from doing more to support the product as a strategic platform for standardizing real-time development? Or from some other source, the proverbial killer application for real-time processing?

Funny reading about CEP and risk analysis

Marc recently wrote a post about how Complex Event Processing (CEP) products are software tools and not a solution to the problem of risk in the capital markets. Although his post was written out of frustration, I found myself laughing (not at Marc, at the content that he analyzes). But I find humor in strange places.

For example, looking at the premise for the article that Marc criticizes:

New complex event processing applications promise to help firms get a better handle on their risk exposure, but can CEP erase Wall Street’s risk management woes?

That question is funny to me because I imagine replacing the three pages of text that follow with the real answer:

No, and that’s a silly question.

This kind of thing doesn’t frustrate me so much because anyone involved in risk management already knows how absurd that question is. If there are executives out there who even consider that software and not strategy is the solution to their RM, they are probably beyond help and should just throw a few bushels of cash on a fire. Same result, much less effort, and at least you could toast some marshmallows.

I suspect that the only people who would take this article seriously would not know risk management if it were standing under a huge, neon “this is risk management” sign. I guess my point is that I’m not very worried about anyone with any influence taking the article seriously. And I hope I’m right about that.

I really do wish that we could get better articles about CEP, though.

If someone were looking to write a real article about risk and CEP, here is a suggestion for the main points:

1. Risk management has many components and some people feel that risk metrics that are updated in real-time will be an important component. Under this heading, we could identify exactly who feels this way, other than CEP vendors.
2. Firms looking to develop real-time risk metrics capability increasingly turn to CEP products to ease the burden of building and maintaining the software. Here we could get some quotes and metrics about how much CEP products can help build real-time software.
3. However, CEP products are software tools that don’t come prebuilt in with risk algorithms. Some vendors provide prebuilt risk products on top of CEP products (list of these vendors goes here). But using a CEP product will, by and large, mean incorporating your own risk algorithms, so make sure that the product can handle them. In any case, the place to start will be with continuously assessing the RM strategy and needs, and if a real-time component is needed then an evaluation of the capabilities of various CEP products and/or prebuilt risk management solutions.

And the caption for the article can be something like:

Will firms adopt real-time metrics as part of their risk management strategy? And how can CEP products help?