Economic data is commonly expressed in aggregate. Consumption and investment behaviors are summed up as Gross Domestic Product, prices of a basket of goods are analyzed to provide an aggregate Consumer Price Index, and Unemployment and tax records are analyzed to provide unemployment rates. Economic concepts are broken out by industry, geography, or other criteria, but economic classifications are still presented in a "top-down" fashion.
Huge strides have been made in the field of data analysis. Techniques for dealing with large data sets have emerged, and "big data" has become a trendy buzzword. Fields which previously relied on intuition are being revolutionized by this new availability of meaningful data.
Economics is a mature field. Econometrics is essentially what data science was before the term data science existed. The theory behind the aggregation of the economy is well developed, and these numbers have been accepted by the business and academic community. New data sources are available, but markets still move on official data releases.
New advances are being made in collecting economic data. Companies like Premise are working to aggregate price level data from the ground up in developing countries, while MIT's Billion Prices Project is collecting this same data by scraping the web.
These still remain two separate worlds however. Economic data was aggregated in the first place because the sort of data collection we have today never existed, and was always meant to be an approximation of the behaviour of the economy.
Change needs to occur on both sides of the divide, and the Bank of England's release of granular survey data can illustrate how this can be bridged. I created this visualization to help move back and forth intelligently between detailed survey data and broad economic concepts, and I hope this can be used to inspire more collaborative innovation between these two fields.
Start by choosing an economic concept below you'd like to visualize.
The Bank of England performs a quarterly survey of the Business community across the UK. Included in these surveys is a set of quantitative scores. These scores range from a low of -5 to a high of 5, and across eleven economic concepts: demand (nominal turnover), exports, investment, capacity utilisation, employment, recruitment conditions, total labour costs per employee, pay (salary per employee), non-labour costs, output prices, and profits (growth in pre-tax profit as share of turnover).
According to their documentation: The Agents' Company Visit Scores, The Bank of England mapped seven of these score concepts to headline economic concepts coming from the Office of National Statistics:
ONS Economic Concept | Company Visit Score Concept |
---|---|
Value Add | Demand (Nominal Turnover) |
Exports | Exports |
Business Investment | Investment |
Total Employment | Employment |
Average Weekly Earnings: Total | Total Labour Costs per employee |
Average Weekly Earnings: Regular Pay | Pay (Salary per employee) |
Gross Operating Surplus | Profits (Growth in pre-tax profit as share of turnover) |
In addition, each of these scores have a current and a future component. Because the economic concepts are time series, the first comparison one could highlight of these scores is to view how average company visit scores have changed over time. In this initial level of analysis, the future scores, when aggregated, should show some insight into how the economic concepts are expected to evolve. Below you can choose either current or future from the drop down menu:
The mean company visit scores, for most concepts, track the percent changes of their corresponding economic concepts. While the future scores give some foresight into how economic concepts will evolve, the real strength of having access to the underlying company scores is to get an idea of the distribution of scores in the economy.
The scores released from the Bank of England include dates and broad industries to identify them. Below, choosing a date from the drop down menu will show a distribution of the scores at that point in time.
In these histograms, you can see how the distribution of company visit scores change over time. In a lot of cases, one can visualize a near normal distribution moving back and forth depending on what its mean score is. Many concepts have a spike at score 0, perhaps highlighting a human preference in the scoring.
As mentioned above, these scores are coded by broad sector. Selecting a sector from the drop down above will update the histogram to display the distribution for that particular industry. In addition, the average company visit score line graph will be updated as well.
This allows someone to seamlessly move across different scales, and see how changes in scales roll up to average behavior. The more differentiated classifications available, the more useful this sort of tool would be. For example, if demographic and geographic data were added to these scores for example, one could filter by the gender of the proprietor to see if women are more successful business owners in certain regions.
Having the granular scores allow us to answer questions that could never be answered with top-down data. Dividing economic data into industries isn't the same as manipulating the underlying data. Once you're able to do that, you can answer more sophisticated questions.
Here for example, one can compare concepts on a firm by firm basis. This is beyond using one economic indicator to predict another in an economic model. Here, one can actually see what happens to a firm with a certain characteristic.
Choose a second concept from the drop down menu below to view a scatter plot across two types of Company Visit Scores.
This graph shows a motion plot of two types of company visit scores. The original concept is on the x-axis and the new concept is on the Y-axis.
The raw data is filtered by the original concept's score (from -5 to 5). For each of these groups, the average of the second score is taken. This graph is animated by time, and the size of the circles correspond to the number of companies with the corresponding original score.
Granular data can be used to answer questions about individual actors in the economy in a way that even the most precisely industry breakdown won't be able to. While this visualization can serve as a model, there is still a lot more work to be done.
I believe economic concepts should be rethought in the age of big data. Economic aggregations as they currently exist can trace their pedigree back to John Maynard Keynes' work following the Great Depression, and even further back to the founders of the field. These mental frameworks existed because people needed a way to express what was happening in an economy with imperfect information. While our information is far from perfect, we're closer than ever to being able to form a complete picture of the economy from the ground up. How important is GDP when the data behind every consumer purchase and investment is available electronically? How important are baskets of goods when the price of everything can be researched online?
Despite this, the economic work done by our predecessors cannot be forgotten. We've only had this enhanced view of our economy for a handful of years. It's extremely useful to perform deep historical dives in order to get a long enough view of the dynamics of human economies. Just as economic data needs to be atomized, the granular data being created needs to be scrutinized, and added to a common framework that includes what we have historically.
It's in this way I hope the best of both worlds can be achieved, and we can combine the wisdom of our predecessors with today's technological advances.