Novel Economic Indicators
By Jason Hines on September 14, 2009
When monitoring large amounts of media there’s an interesting opportunity to track potential economic indicators – direct and indirect – such as bankruptcies, corporate insiders buying/selling shares, credit ratings, analyst ratings, etc. at large scale. In this blog entry we’ll explore bankruptcies and ratings data from Recorded Future news analytics to make a high level judgement of whether the economy has turned around or not.
Bankruptcies are certainly an interesting part of the economy. Clearly when they start to happen at large scale things are about to go wrong (or perhaps already are). The ripple effects of single bankruptcies (General Motors, Lehman Brothers) can be massive across supply chains.
In this case we’ve reviewed the count of unique bankruptcies per month in aggregate as well as a simple rate of change metric for the same. The unique count is important as a bankruptcy event clearly can be covered repeatedly over and over – e.g. the General Motors event flooded the media stream in May/June. In the Spotfire visualization below we visualize the count of unique bankruptcies (red line) as well as the rate of change (blue line) over 2009 and can certainly see that bankruptcies picked up across the year, and peaked in July. The rate of change peaked earlier and has already turned negative. So, it seems that perhaps things are improving here!
Looking at a simple treemap of the bankruptcy events – organized by industry (just one level hierarchy) – we can see on the left in the visualization below that the largest sector for bankruptcy events clearly is financial services, perhaps not a surprise.
As a precursor to bankruptcies we may want to look at credit rating events – which we also pick up in Recorded Future. Reviewing those events below, across the year, in a somewhat simplified/filtered form, we can see how downgrades peaked earlier in the year, affirmations seems to be on the rise, but actual increases in credit ratings are still flat. The time lag between bankruptcies and credit ratings is quite logical – and would be interesting to model.
Finally in reviewing analyst ratings for 2009 we clearly see an increase across the year in upgrades – and actually perhaps now they (upgrades) are starting to flatten out (quite logical) – providing us with a sense of that analyst ratings first change, followed by credit ratings, before finally bankruptcies occur (yes, of course there is much more to story than that!).
This type of data could potentially be used to create “macro indicators” – providing perhaps an early glimpse/prediction of what a later “official number” will be – which could be highly interesting in for example gaining an “up on the market” in terms of early indications of a impactful macro indicator. To improve it many things could be done – perhaps only looking at a particular industry or a specific data source (e.g. SEC filings).
However – the same sort of data could also be used to look at a particular company or portfolio of companies – what about if you could form an alert which allowed you to track the chain of a cut in analyst rating, a cut in credit rating, and first chatter of bankruptcy (and other events of course!) for a large set of of names/companies – and keep you notified along the way? Certainly something we want to do with Recorded Future.
There are many ways this analysis may be argued as too simplistic – i.e. too short time period, should be compared against a baseline, seasonal adjustments, does it only look like an improvement – in reality the media flow just slowed down in August, or for that sake – media just getting bored with reporting bankruptcies. All these can be battled though – more/longer data frames, focusing on particular sources (SEC 8-K filings for bankruptcies), etc. etc.
The point of the exercise is less about trying to claim that this data is more accurate than for example an aggregation of bankruptcy court filings from primary sources or a number you may receive from an analyst – but more about how you can a) shape your own indicators from a novel data stream (media/internet) and b) be able to judge/read/react to those indicators faster than you can when waiting for a number that everybody else is waiting for (e.g. compare housing/jobless/etc. numbers).
Even more interesting is of course to combine this data with more indicators extracted from primary sources, internet, and media – insiders buying/selling shares, company restructurings, layoffs, buybacks, IPOs, company expansions, etc. Recorded Future has all those event types and many more – and you can combine them – e.g. into a 4 step complex event that in total would indicate that a company is “turning around” or “flatlining”. Lots of opportunities for unique insights!
Finally: we’re obviously going to do some of the above improvements – and also look at incorporating forecasting/trending techniques here – as well as comparing results to 3rd party numbers/indicators.
As always, we welcome your comments below!