Predicting and Controlling HyperInflation in Venezuela

In a prior post (here) I presented one measure of Inflation in Venezuela (NY.GDP.DEFL.KD.ZG from the World Development Indicators) with an attractor path and 98% Bootstrap Prediction Intervals.**

Notes

** Inflation is usually presented as percentage changes in the GDP Deflator. For technical reasons, I do not use the conventional measure in statistical estimation. Change scores, Q(t) + E(t) - Q(t-1) - E(T-1), contain error, E. In time series data, the errors are unlikely to cancel and are typically not normally distributed and will not cancel in probability. The regression coefficient, B,  for a time series model of the deflator, DEFL(t) = a + B DEFL(t-1) + E(t-1) is better estimate of deflation and is free of error,  E(t-1).

The Taylor Rule and US Economic History

 

In a previous post (here), I present four regimes for Automating Fed decision making. Three of the regimes are designed to control the Federal Funds Rate (FFR) during crisis. All four regimes need to be compared to the Taylor Rule, a single equation loss function for controlling the FFR (see graphic above) and other proposed rules for the FFR.

The tracking for the FFR (red line in the graphic above) and the Taylor Rule (blue line) is pretty good but not perfect, particularly in the periods after crises (1950-55, 1970-75, 1980-1990, 200-2010 and 2020 to the present).

Notes

FOMC Measurement Model

FOMC AIC

Notes

FED Index

FED Structural Model

Trump vs. the Federal Reserve: Part II

 

In a prior post (here), I concluded that Both Trump II and the Federal Reserve (the Fed) are right: Interest rates are too high and they can't be lowered given the economic uncertainty of Trump II policies. So, from the graphic of the Federal Funds Rate (FFR, the Fed's primary policy tool), interest rates need to get down to the attractor path from the USL20 model, but we are still way above the 98% prediction interval for the attractor path. From the perspective of Systems Theory, the attractor path graph brings up the question of how good a controller has the FED been historically?

From the perspective of Cybernetics, which likes to think in terms of control bands, the Fed has been an Erratic-Controller going sharply outside the 98% predication intervals for most of the Late 20th Century. Prior to 1970, manipulation of the Fed Funds Rate (FFR) had little impact.

There some very interesting departures from the Attractor path during the Late and Long 20th Century. Essentially, the Fed was an integral part of history during the period (see the Notes and the Readings): (1) The Vietnam War, (2) The Arab Oil Embargo, (3) The Great Inflation and the Volker Deflation, (4) The Great Moderation, (5) The COVID-19 Pandemic and (6) The Rebound (Post-COVID Inflation).

In order to evaluate FED performance, we need a model for predicting the Fed Funds Rate (FFR). I have three models, one driven by the USL20 model, another driven by the WL20 model and a Business-As-Usual (BAU) model. The USL20 model is a strong model (using the Akaike Information Criterion, AIC) but the model is unstable (I'll comment on that below) so I have used the WL20 model as input because it is stable and provides an interesting counterfactual path. The attractor path (conducting a free simulation starting at 1950) shows two periods: (1) The Pre-Volker Deflation (before 1980) and (2) The Modern Fed after 1980. 

In the Vietnam- and the Oil Crisis-periods, the attractor path suggests that the Fed needed to do more, increasing the FFR to stop Inflation before it started. After 1980, the Fed was responding to World-System signals, manipulating the FFR to deal with crisis (a little late in 1990, during the Dot-Com Bubble).

In spite of reasonable policy prescriptions from the USL20 model and  the WL20 model, the BAU model is still best given the AIC statistics. The model suggests agin (graphic above) that the Vietnam-period needed higher interest rates, but after that the model calls for a steady state FFR path.

What are we to make from the predictions of these three FFR models and the History of the Late (and Long) Twentieth Century):

• Fed policy actions (at least the FFR) are erratic. All the models call for smoother policy responses. For political reasons, the Fed may feel it has to make strong (rather than gradual) statements.
• The Fed seems to have found it's footing after the Volker Deflation, but then had to improvise through the Subprime Mortgage Crisis.
• If the US Economy and the World System are reaching a Steady State, where there is strong political pressure to lower interest rates an stimulate growth (even though rates are close to the Zero Lower Bound, ZLB), the FED may be forced to use the BAU (or even a Random Walk, RW) model to avoid political pressure.

Do we have a good idea what drives Fed policy? I'm not sure. However, there is one more "Structural" model we can look at (AIC = -455.7) that creates a Cybernetic Reaction function for the Fed mandate and uses it to drive policy. I"ll look at that more in another post (here).

For the present, the future of Fed policy actions will be hard to predict but we can continue studying counterfactual alternatives and watch what happens in the real world. 

Notes

• The Vietnam War (1965-1973)
  
  
  
  
  
• The Arab Oil Embargo (1973-1974) 
  
  
  
  
  
• The Volker Deflation (1979) 
  
  
  
  
  
• The Great Moderation (1985-2005)
  
  

  

  
  
• COVID-19 Pandemic (2020)
  
  
  
  
  
• COVID Rebound (2023)
  
  
  

Readings

• The FED Overview: History of the Federal Reserve
• The Collector The Economic Effects of the Vietnam War
• NY Times What Vietnam Did to the American Economy
• Columbia SIPA The 1973 Oil Crisis: Three Crises in One--and the Lessons for Today
• INET Oil and the Energy Crisis: A Reanalysis
• FRB The Great Inflation and the Volker Disinflation
• St. Louis Fed The Volker Tightening Cycle
• The Fed The Great Moderation: 1982-2007
• The Fed Monetary Policy and the Housing Bubble
• The Fed The Great Recession December 2007 - June 2009
• The Fed The Subprime Mortgage Crisis 2007 - 2010
• US Treasury COVID Economic Relief
• Brookings The US Recovery from COVID-19
   

USL20 FFR 2026 Model

USL20 Measurement Model

WL20 FFR 2026 Model

WL20 Measurement Model

FFR Models AI Statistics

FFR Model BAU

Project 2025 and the US Federal Reserve

 

It is probably a fair summary to say that Project 2025, the Conservative Manifesto for change in US Federal Agencies, wants to blow up the Federal Reserve in addition to a number of other agencies. ChatGPT provides the following summary:

The graphic above is a side-by-side comparison of how Today's Fed functions as compared to the recommendations of Project 2025. The recommendations are that (1) the Fed should concentrate only on Inflation (screw the Unemployed Workers), (2) not be a lender of last resort during Bank panics, (3) Eliminate the Fed purchase of non-performing assets from Bank and Corporation Balance sheets, (4) Limited only by Free Market discipline rather than regulation and (5) Create a Free Banking System that prints it's own money and is free of regulation.

Personally, I love the last recommendation: the worth of your money would be determined by the reputation of the bank that issued the money, e.g. Goldman Sachs money would have a value different from J P Morgan money and would fluctuate over time. 

Chapter 24 (and indeed the entire Manifesto) also has (1) no data supporting the recommendations, (2) no modeling of the counterfactual world in which there was no Central Bank and (3)  unbalanced arguments supporting it's extreme proposals. Luckily, ChatGPT warns:

For more of my posts on the Federal Reserve with data and modeling, see the Blog Roll.

Why does the US Fed No Longer Try to Control the Money Supply?

 

Milton Friedman famously argue in a Monetary History of the United States that the US Fed's failure to control the money supply led to the Great Depression. Chat GPT notes:

So, it should be somewhat confusing for Conservatives to the learn that, after the Volker Deflation, the Fed stopped monitoring the money supply. So (1) What Gives? and (2) If it wasn't the Money Supply, What Caused the Great Depression? ChatGPT again:

And, the Cybernetic System that describes the Fed (ChatGPT) is:

Just to check this line of reasoning, I've re-estimated the Fed Reaction function (from this post in the Notes below) this time including the Money Supply (M1 and M2). A few things to notice are (1) M1 and M2 are about equally weighted in FED1, the overall growth index. (2) M2 appears with smaller weights in the Unemployment Controller (FED2) and the BANK1 Controller (FED3) and (3) the M1-M2 controller, FED4, explains very little variance (less the 0.01%).

Over time, in the graphic above, FED4 is relatively constant with a small dip during the 2010 Great Recession, when the Fed was very active.

Notes

Expanded Fed Reaction Function Controller

The basic Fed Reaction function controller is presented in this post and does not include the Money Supply.

A Model for Policy Reform and Automation of Federal Agencies

 

The Trump II Administration's attempt to destroy Federal Agencies (Department of Education, Department of Energy, Environmental Protection Agency, USAID, etc., etc.) brings up the questions of (1) What are all the Federal Agencies actually doing, (2) How do citizens evaluate what they are doing and (3) How do citizens have some input on policy decisions (current voting does not provide enough control and feedback). Intelligent policy reform and automation could help start the process of evaluating Federal Agencies rather than waiting for Presidential wrecking balls.

The graphic above (with some modifications) was generated by ChatGPT in response to the question of whether the US Federal Reserve could be automated. It can be generalized to cover any Federal Agency. In Detail:

• MANDATE Laws passed by Congress define a REACTION FUNCTION that must be minimized.
• DATA The Federal Government currently gathers all the data need for decision making.
• VALIDATION Real time is noisy and has to be validated.
• MULTI MODEL There is no one perfect model of the economy (what ChatGPT calls the "Big Brain"). The multi-model, ensemble approach to forecasting is also used by Hurricane forecasting (see the Boiler Plate). I would recommend using the Akaike Information Criterion (AIC) to evaluate models.
• REGIME CLASS There are different regimes that the agencies operate under in addition to Business-as-Usual (BAU). The different regimes have to be defined in the REACTION FUNCTION.
  
• POLICY ENGINE The Policy Engine would present multiple models predicting actions under different regimes.
• REACTION FUNCTION  would describe the best action under different regimes (here). It describes desired states (Q*) to be compared with an actual states (Q). (Q*-Q) is a multivariable Cybernetic Reaction Function used to control systems.
• VOTING I have allowed for citizen voting using a cell phone app to provide democratic input similar to the Atlanta Fed EconomyNow App (see below in the Notes) but with voting input.
• UNCERTAINTY The main purpose of the Agency Board is to allow human input, accountability and avoid technocratic mistakes under uncertainty. 
• OVERSIGHT Oversight involves identifying regimes that are not BAU.
  
• OVERRIDE Under conditions of uncertainty, the  Agency Board can override the Policy Engine.
• POLICY Ultimately, the Agency decides policy using all available information.
• COMMUNICATE The  Agency Board is charged with transparently communicating the policy decision and explaining their evaluation of existing data and output from the Policy Engine. The communication would include open evaluation of how well the agency has performed under different policy regimes.

So, ultimately and obviously, the Agency Board cannot and should not be replaced by a computer. However, a process can be set up that uses all available multi-model computer input and public voting to allow the Agency Board to make transparent policy decisions.

ChatGPT reports that:

US Executive Agencies (the Department of Defense, HHS, Homeland Security, Dept. of State, Immigration, the White House, Congress, the Supreme Court, etc., etc.) have all been compromised by poor leadership under an authoritarian political regime (the Trump II Administration). Democratic  voting itself has been under threat and is too slow and indirect a process to prevent abuse. Indeed, the entire system seems quite Anachronistic.

The model above provides a general approach to reforming failing institutions. The US Federal Reserve (here) could be used as a model.  

Atlanta FED EconomyNow App

 

The Atlanta Fed EconomyNow App would be extended to allow voting on policy options, as another source of input for the Fed.

Four Regimes for FedOS 1.0

In prior posts (see the Notes below), I have explored the question of whether the US Federal Reserve could benefit from more automation and transparency. In this post I look at the question of whether automation could help identify different Crises Regimes in addition to Business-as-Usual (BAU, where automation is likely to work quite well).

Using Principal Components Analysis (PCA, see the Notes below and the Boiler Plate) I have identified a Reaction Function (FED) for each of the four scenarios: FED1 = BAU, FED2 = Full-Employment GDP Controller, FED3 = Financial-Inflation Controller and FED4 = Inflation Controller. The Control Functions are plotted above.** Over time: FED1 (BAU) increases but hits a peak in 2010, FED2  shows numerous cyclical peaks, the largest being after 1980, FED3 is relatively stable but peaks around 2010 and FED4 is relatively stable without peaks.

The next question is "...what action the Fed has taken historically during shocks". The Shock Decomposition diagram above (created by the FED model in the Notes) shows two regimes in which the Fed raises the Fed Funds Rate (FFR), Business as Usual and Inflation, and two regimes in which the FFR is decreased, Unemployment and  Financial Inflation. In general, these crises happened at different times historically so the Fed was not conflicted about policy directions.

Another point to notice from the Shock Decomposition is that the FFR effects are small except for the Inflation controller. And, in another post I have show that FFR changes have small effects on the economy (here).

These results do not mean that the Federal Reserve is ineffective. A major function of the Federal Reserve to regulate the US Banking System. Manipulating the FFR is one policy measure among many that are important to a stable Banking System. The Fed saved the Banking System during the Subprime Mortgage Crisis (although it can be argued that stronger Banking Regulation could have prevented the crisis in the first place). Without the Fed, economic consequences would have been much worse.

Notes

• Could the Fed be replaced by a Computer Program? No, but more automation, communication, public input and transparency would help (and could be a model for other Federal Agency reform).

Reaction Function Index

** The Reaction Functions, plotted at the beginning of this post, are independent (by construction). Notice that the FED2 function (Full-Employment GDP Controller) does a very good job of tracking Crises in Late 20th Century US Economic History:

• Great Inflation Google AI reports that "Inflation in the 1970s was caused by a mix of supply shocks (especially oil crises from OPEC), expansionary fiscal policies (Vietnam War, Great Society spending), and monetary policy errors, leading to "stagflation" (high inflation and unemployment)".
• Dot Com Bubble The tech–media–telecom (TMT) bubble and low interest rates led to a speculative frenzy on Wall Street.
• Subprime Mortgage Crisis A multinational financial crisis that occurred between 2007 and 2010, contributing to the 2008 financial crisis and required the Fed to invent new programs for dealing with the fallout. 
• COVID-19 Pandemic Caused "severe social and economic disruption" and required Fed Action.

All these crises require Fed policy action and the FED2 index captures the full extent of each crisis.

Control Functions with independent Regimes and empirical weightings can be created using Principal Components Analysis (PCA) on standardized data (see the Boiler Plate): FED1 = (Overall Growth in the Indicators), FED2 = (0.9736 LU - 0.206 GDP) Full-Employment GDP Controller, FED3 = (0.8534 BANK1 - 0.381 GDP - 0.3165) Financial-Inflation Controller, and FED4 = (0.7247 P.GDP. - 0.682 GDP) Inflation Controller.

Banking Index

An index of Banking Stability can be created using the following indicators: FB.BNK.CAPA.ZS = Bank Capital to Asset ratio, FB.AST.NPER.ZS = Nonperforming Loans, and FB.CBK.BRCH.PS = Bank Branches.  BANK1 = (Overall Growth with relatively equal weightings), BANK2 = (0.847 FB.AST.NPER.ZS - 0.462 FB.CBK.BRCH.PS - 0.261 FB.BNK.CAPA.ZS) Nonperforming Loan Controller and BANK3 =  ( 0.743 FB.BNK.CAPA.ZS - 0.657 FB.CBK.BRCH.PS) Branch Bank Capitalization Controller.

 

Over time, in the graphic above, the BANK index captures the Dot-com Bubble, the Subprime Mortgage Crisis and the resulting Dodd-Frank Legislation.

It is interesting that the FED4 Inflation Controller explains very little variance as a policy regime (Less than 0.05%, graphic above). Given that Inflation Hawks are constantly warning that Fed actions are inflationary, the historical record does not show FED4 as an important Crisis Controller.

ChatGPT list somewhat different non-BAU crises and suggests other policy responses:

Establishing policy actions to take during new and novel financial crises is an important and continuing responsibility of the Fed the the Fed Open Market Committee (FOMC).

FED Model

The state space mode above was used to generate the Shock Decomposition in the text. The model is nonlinear and stable.