Is Musk right that the Fed's data 'has too much latency'?
Does the Fed operate with data that has ‘too much latency’?
Elon Musk’s Twitter bombs are often entertaining, but he is not always completely wrong. For instance, on the morning of 27/4 Musk tweeted ‘the data with which the Federal Reserve is making decisions has too much latency’.
The Federal Reserve is the name for the US central bank, the body that decides on the level of interest rates [and lately quantitative easing] in order to control inflation, smooth the booms and busts in the real economy, and look after the stability of banks. Musk is claiming that the data that the Fed is using is too old and out of date. Musk is using explicitly technical language here, when it could be stated more simply, though with less pzazz, as ‘the Fed’s data is old.’
He was even clearer on 30 April:
Having up to date information for monetary policy is super important. Imagine trying to drive a car where you saw not what was actually happening, but a 360 video of what was happening a few seconds ago. Getting to your destination safely would be much harder and less likely to achieve without an accident, although not impossible.
Actually this is not a super clean example. Light takes time to travel from objects in the road, into the light sensitive cells in our retinas, and then their responses take time to travel into the area at the back of the brain where visual information is processed, and for it then to be processed and interpreted by other bits of the brain, and so on. So we do drive with road data that is out of date [‘has latency’]! However, it would be a whole lot more difficult if the lag was made longer.
Rocket flight is an example Musk’s engineers at SpaceX will know all about. One of SpaceX’s big achievements has been to land rocket boosters back on solid ground and even on a raft floating in the ocean. These descents require a constant process of data collection on the rocket, monitoring, applying corrective forces using changes in the orientation and power of the engines, collecting more data and so on. By the time the data is collected and processed and corrective action applied, it will ‘have too much latency’, ie no longer be true, because the rocket will have moved some more. This has to be factored in to the control process. Using, theory, trial and error, and computer simulations of re-landing a rocket, the engineers learn how to devise control rules that allow for latency in data on what the rocket is doing.
Likewise latency or lags complicate monetary policy. If you can’t see and react to a shock quickly, then by the time you do see what happened, it may be too late to do anything about it.
Musk tweets have to be interpreted in the context of his broader interests. He seems to relish baiting US and world institutions, particularly those that define the liberal order. He has a thing for crypto, as the brief appearance of the dogecoin logo on the Twitter homescreen reminded us. Recall how he got Tesla to organise accepting payment for its cars via Dogecoin; and the $140m loss Tesla made on its Bitcoin investments in 2022. Trolling the Fed, guardian of fiat currency, is boilerplate for crypto enthusiasts.
Most pertinently of all, Musk’s $44bn purchase of Twitter was financed with a lot of debt [about $13bn]. Interest rate increases raise the cost to Musk/Twitter of servicing that debt. They increase his outgoings at a time when Twitter has been struggling with a plunge in advertising revenue [as blue chip companies disliked the brand risks they saw coming with Musk’s new content moderation policies] and the operational difficulties caused by deep cuts to the Twitter wage bill. Musk wants to head off interest rate rises, and claims that they are not necessary. The recession that will bring inflation down, he says, ‘is already here’.
Leaving aside that he might well be just talking to his own financial interests, is it anyway true that the Fed’s data has ‘too much latency’? What does this claim mean?
The latency of the data the Fed - similar to central banks in the other rich economies - varies. Policy makers will have minute by minute information on financial prices; bond yields, stock indices, spreads, foreign exchange rates. The data most central to its objectives, inflation, arrives with a lag. As I write this we are at the end of April 2023 and the latest data is for inflation up to March. March is also the latest date for changes in employment, unemployment, producer prices and hourly earnings. Changes in GDP also very important for the Fed, is a little more out of date. There is an ‘advanced estimate’ for the first quarter of 2023 [Jan-March]; the latest, revised estimate is for 2022Q4 [September-December]. GPD data illustrate a common feature of much economic data, that the less latency it has, the lower its quality.
Is this ‘too much latency’? Musk quips about his own data, which he is implying is better, both in volume and in its lack of latency:
It’s hard to argue that more up to date data would not be better. But acquiring it could be very costly. It would require a bigger budget for the Federal Reserve System. It would also require a bigger budget for bodies like the Bureau of Labour Statistics which collects a lot of it. For data that is done with surveys, carrying them out faster would require more people.
At some point the desire to eliminate latency also starts to generate intolerable costs for those whom you are collecting data about. For example, having continuous real time data on everything would require either Federal agency access to your company’s bank account and internal financial monitoring tools, and your personal bank accounts, intrusions that many would find unacceptable. Or you would be continually bombarded with survey questions about your incomings and outgoings.
Would it nevertheless be worth more money and the intrusiveness of collecting more data? If so, what would be the optimal amount to collect before the benefits tail off? I have no idea. I love data. I find it fascinating. It was partly what drew me into a life in economics. I find it hard to set aside my own interests and address the question in sober, utilitarian terms. For people like me, more data would be great. But that does not mean it would be good way to spend our collective resources getting more of it.
One reading of Musk’s tweet is ‘the Fed should be using the data I have; or also using the data I have, which has less latency’. Musk has access to his own companies’ real time data. [Although even that will have some latency, as staff focus on doing things, rather than continuously reporting what they are doing]. But he and other similar CEO’s would not want to share it with the authorities, for perfectly ok reasons. We value our privacy and fear that data will be handled poorly, seen by our adversaries, or used for reasons we did not sign up to.
Musk overstates the usefulness of his data, though. Although it is fresh, Twitter/Tesla/Starlight/SpaceX data provides a very noisy and highly unrepresentative signal on data for the whole economy, which is what the Fed has to be concerned with. The US or any other economy is not much like a conglomerate that manufactures EVs, rockets and internet infrastructure.
It will have some signal for the rest of the economy, but it will be weak, and requires that you spend a lot of money on the economics and statistics to work it out based on relationships in the past, money that we can bet that Musk has not spent.
Musk may well also have economically relevant data derived from tweets on his platform. Readouts on the frequency of words like ‘inflation’, ‘uncertainty’, ‘recession’, ‘crash’ or ‘bank run’. But the signal that this tweeting has for the Fed’s policy objectives will take some effort and expertise to extract, tasks that the Fed is probably much better at than he or his staff. Central banks and private sector central bank watchers and academics have started collecting this data. My sense is that at the moment it’s just a bit of fun and does not have any demonstrable benefit over data they already have.
In fact the Fed and the agencies it depends on for data do something a bit like what Musk implies he is attempting, only I think much, much better. The Fed’s data has different degrees of latency. Some of it is more out of date than others. Some data get revised. Early estimates are based on partial samples, and attempts to extrapolate from the older, missing data. The Fed uses economics and statistics, the relationships in older, more accurate and more complete data to extract the signal out of newer, noisier data, or fill in the data frontier. Economists, statisticians and engineers call this ‘filtering’. Using a lot of observables, new and old, and some theory about how the underlying unobservable true quantities of interest evolve, to extract the signal from the noise. Models not dissimilar to this - which use the Kalman Filter or similar - will have been programmed up by Musk’s own engineers in SpaceX.
The Fed, like any central bank, is surely not doing its job perfectly, even given its limited budget. It is staffed full of humans who make mistakes and have their own interests - pursuing their public profiles, their private research interests - aside from the job they have been given by statute, delivering price stability and full employment. The disciplines their staff hail from are also not perfect; the macro-economy is still poorly understood and techniques for diagnosing its laws of motion and forecasting are evolving, and subject to fads and orthodoxies. They are anchored, if only loosely, by competition on the field to figure out new and better methods and publish them; and by the fear of being embarrassed by policy mistakes and leaving a legacy of incompetence. But could the Fed’s leadership do anything concrete to improve? Maybe, though I doubt it could do anything transformative. Could legislation be passed to reform the institutional framework that gave rise to a better leadership to improve these things? Not in the current political climate.
Musk is therefore not necessarily out of line for suggesting that the Fed may be in the process of messing up. There is a large community of ‘Fedwatchers’, economists and market experts watching the day to day behaviour of the Fed - doing exactly this. They make their money selling analysis and opinions. If they are right about what the Fed should be doing, chances are the Fed will follow suit at some point, and those trading stocks and bonds can figure out what that will do to financial prices and react accordingly. ‘The Fed is overdoing it with interest rate rises and that will generate a big recession’ is a forecast of what will happen and implied prediction of how the Fed will later correct, and this implies an investment strategy. So you will get a lot of professionals making Musk-like comments all the time.
In fact, the Fed itself will - if it’s like other central banks [I worked in the UK central bank and speak from experience] - be turning commentaries like this into data that it monitors and uses in the policy process. Because of Musk’s obvious financial conflict of interest, and the clear lack of utility in the data he sits on that he keeps to himself, his comments will be easy to ignore. But the regular financial analysts won’t be ignored. The FOMC will stare often at charts of where their own forecasts of inflation and activity are relative to those of outsiders and what economists outside think they will do to rates. One kind of data the Fed has that is abundant and has little latency is Musk-like comments about itself.
After all that, and setting aside his claims about the latency of the Fed’s data, is Musk right? Is the Fed about to overdo it with interest rate rises and generate an unnecessarily large recession? Inflation seems to have peaked, after all. Will it not return to the 2% target without any further rises in interest rates? This question really deserves to be answered in a post all of its own.