Dow Jones has begun publication of a new Economic Sentiment Index, which is based on a text mining analysis of five million news articles referencing the U.S. economy since 1990, rating words such as “recession” and “depression” as negative and “recovery” and “rebound” as positive.
This index doesn’t give us new information on the economy itself, only what reporters are saying about it. Just because lots of people say the same thing doesn’t make it so, unless there’s merit to (1) the wisdom of crowds, or (2) self-fulfilling economic expectations. But in any case, it’s an interesting series.
why does the plot call a local extremum an inflection point? aren’t inflection points where the second derivative changes signs?
Good point, anonymous. I simply reproduced the graph from Dow Jones, and would not myself have labeled these as “inflection points”.
Heh, inflection point appears to mean more than merely “local extremum” but instead “local extremum which happens to correspond to an appropriate external event”
There are plenty of extremums on the graph which are not so labeled, and it’s not clear what mathematical distinction you’d make to rule them out versus the ones chosen.
FWIW: I tried a much smaller experiment (a few words, sources, time), but concluded that my word index was not useful for making real time decisions on some future economic events of personal interest. Could well be my fault, but thought I’d report the experiment and negative results from using words from the mass press. My conclusion is that reporters, as a large group, are not ahead of events. Only a few commentators are ahead of events. The DJ sentiment index has shapes that yell failure to me for forecasting purposes. For example, the 2001 recession area has 2 “false bottoms”, and then the series drops further, and the recession continues. What can we infer from the index in the 2008 area with that prior index behavior? In my experience, good predictors have different shape behavior. Again – FWIW.
Readers may be interested in a similar exercise that is carried out weekly at Knowing and Making – summarising the words used in all economics blogs postings of that week.
Click for
this week’s analysis or here for a list of the historical clouds generated to date.
We’re not as comprehensive as Dow Jones’s index of course, but quite interesting to see the difference in concerns expressed by economics bloggers compared with general news journalists.
I would say that the use of ‘inflection point’ term instead of ‘local extremum’ one might be revealing the way the index is calculated. If the series is a ‘slope’, rate of change,…(?) of some other series, then the use of the ‘inflection point’ would be justified.
Maybe it could work as counting an amount of change of a fraction of ‘positive’ and ‘negative’ words’ appearances between consecutive weeks/days? Smoothed as well, as it seems to have a small lag.
This graph is a good reflection of increased reach of internet and following force of self fulfilling economic expectations whose impact on real aconomy has been amplified allmost double due to Internet.
No wonder recovery starts in China whose information control policy makes many people ignorant of “looming” dangers and thus more resistant against falling into “informed” depressive panic.
The severity of this crisis is in my opinion 100% attributable of information spreading technology, but so is also recovery.
It would be good to have some integrating circuit with reasonable delay /averaging time (1 month?) on all this real time flow of bad/good news. People want bad news ( about others) though. Engineers in electronics have long found out that having a lot of high frequency noise passing thorugh the system does not add to the quality of music, on the contrary.
The Keynesian revolution, which actually started before Keynes General Theory, began to consider aggregate price levels rather than micro prices. Hayek has observed that prices rise and fall naturally in a normal economy and at times this may be seen as a general change in prices. He went on to observe that the general price level in this situation must be allowed to fall. A policy of price level stability in such a condition is as damaging as that of an increase in prices in a normal economy. If the monetary authority reacts to such a price fall by injections of money the problem is transformed into a situation of perpetual and increasing increases in monetary expenditure resulting in the “stagflationist” dilemma.
To us this is well known because of the events of the Jimmy Carter era and stagflation, but Hayek wrote this in 1935 concerning the period from 1927-1932.
Quote:
… up to 1927, I should, indeed, have expected that because, during the preceding boom period, prices did not rise but rather tended to fall the subsequent depression would be very mild. But, as is well known, in that year an entirely unprecedented action was taken by the American monetary authorities, which makes it impossible to compare the effects of the boom on the subsequent depression with any previous experience. The authorities succeeded by means of an easy money policy, inaugurated as soon as the symptoms of an impending reaction were noticed, in prolonging the boom for two years beyond what would otherwise have been its natural end. And when the crisis finally occurred, for almost two more years deliberate attempts were made to prevent, by all conceivable means, the normal process of liquidation. It seems to me that these facts have had a far greater influence on the character of the depression than the developments up to 1927, which, from all we know, might instead have led to a comparatively mild depression in and after 1927. Prices and Production, London: Routledge, 2nd ed. 1935, pp 161-62
So what does this say about our current crisis and more importantly recovery? While the rhetoric out of Washington DC has been about supporting home prices and the pundits are still searching for a “housing recovery,” meaning rising home prices, the actual actions taken by the government have been more toward bailouts of their Wall Street buddies than “price stability.”
This may actually be a blessing. Recovery will come when prices find their level so interventions by the government to artifically prop up prices simply prolongs this process and makes it more painful. But while crony capitalist injections of cash into the pockets of Wall Street shysters may be immoral, the negative impacts on the economy as a whole may be limited. Liberal greed may actually override their propensity to destroy – to coin a Keynesianesque expression.
Im one of the people who helped to create the ESI. Please don’t get too distracted by the term inflection point , it’s in the publicity material which was meant for a general audience and isnt meant in the mathematical sense of a change in the sign of the second derivative. I thought it had been removed, but it seems to have bindweed-like properties. Here it just takes the colloquial meaning of turning point.
As for what the ESI does, the intention is to capture current sentiment about the economy. In theory, thats the sort of information you ought to be able to distil from newspapers.
OK, you might say, but so what? We know the present. Tell us about the future.
Well, Im not entirely sure the various existing surveys and data tell us enough about what’s happening right now. If they did, why would professional economists (not to mention the NBER) often miss the fact that the economy is in recession for many months after it has started?
Based on backward looking studies, the ESI seems to be a pretty good coincident indicator of whats happening in the economy. Interestingly, it even offers a lead on nonfarm payrolls (albeit it works better on the revised series). That could be because employment lags the cycle. But if employment is such a lagging indicator, I dont know why markets often show such strong reactions to the release of the weekly and monthly data.
Like any indicator, the ESI gives false signals. And at times it may lag events slightly. So you might not want to base all of your judgements on it. But assuming it works in future as well as it has done retrospectively – it adds incrementally to whats already available.
Were hoping people see enough in the ESI not just to use the indicator but also to run their own analyses of Dow Jones Factiva newspaper database.
Would a word (count) search for bananas portend inflation??
Some infers news is almost by definition bad news as the American proverb goes bad news travels fast. So I guess the index by its formula negative news biased . “recession” and “depression” comes quicker in the news than “recovery” and “rebound”. I think a minor adjustment is required. How ever its a good work and rough estimation of public sentiment can be assessed effectively.