Do Interest Rates Tackle Inflation?
If correlation means causality then possibly not. It may have an effect, but the effect might be weak on inflation and brutal on society
Contents
Too Long Didn’t Read
Motivation
The Test Question
Method — What’s wrong with VAR models?
Results & Discussion
Closing Thoughts
Too long, didn’t read
As a data scientist and an outsider to the field of Economics, I explored the link between UK interest rates and inflation.
I identified correlations that broadly tie in with models from the International Monetary Fund (IMF) but identified issues that led me to question conventional wisdom in the field:
- Asking a different question: Mainstream economists use ‘VAR models’ in order to reason about the economy whilst making a big assumption that interest rates do indeed tame inflation. This is why I took a step back and correlated interest rate hikes with real-world inflation outcomes for each consecutive financial quarter instead
- Counter-intuitive outcomes: It is plausible that interest rate hikes stoke inflation higher, especially in the earlier stages of interest rate hikes. If it does eventually bring down inflation, the overall effect might be weak
- Major changes to inflation: I cannot see how interest rate changes explain major changes to inflation rate outcomes — I think that other factors probably cause big drops in inflation
- Self-critical feedback: At the end of this article, I list what I think could improve my own work as well as ideas to mix up the field of Economics
The Bank of England head Andrew Bailey recently said that interest rates were working to bring down inflation, a statement that former shadow chancellor Ed Balls showed skepticism on, stating that the US economy is much more influential on our inflation than UK interest rates. I can believe this, so is it necessary to drag down a perilous UK economy with interest rate hikes?
I am a programmer too, just give me the code
For any fellow Python programmers out there, my research code is written in Python and open and available on Github with data sources and statistical results inline.
Motivation
Feeling inspired by another great Bubble Trouble podcast by the brilliant Will Page and Richard Kramer, I decided to take a look behind a sceptical dig at interest rate policies by Will, who is former chief economist at Spotify. He was making the point that a correlation of interest rates and inflation does not mean that they are causally linked.
However, there are ways to help tease out causality in statistics and I wanted to explore this. I grew a big itch, so big that it managed to pull me away from watching every Harry Potter back to back on Netflix and in to the misty world of wand waving Economists.
The Test Question
Central banks claim that raising interest rates helps tackle inflation. If true, then I would expect yearly hikes to the interest rate to consistently correlate with falls in the inflation rate at later points in time.
Rationale
When the Bank of England raised interest rates by 3% last year, it corresponded to an increase in mortgage payments by £500 per year for an average UK household. This means millions of us have £500 less to spend on goods — for most, that’s not something you find behind the back of a sofa. This is one supportive theory behind reducing inflation.
From the description above, I think it should be clear not to look at the interest rate and inflation rate itself, but the difference in interest rate corresponding to a difference in spare household income and, therefore, the difference to inflation rate after a period has passed.
Method (and what’s wrong with VAR Models that Economist use?)
I used lending rates data for central banks and the Consumer Price Index (CPI) data as a proxy for inflation from the Bank of International Settlements (BIS). I used linear regression to test yearly changes to interest rates vs differences to the quarterly inflation rate at each consecutive financial quarter.
I used all UK data from 1960 but excluded the period after 2008 because in my view that period already undermines the use of interest rates.
Why didn’t you use a VAR model like other Economists?
Yes, I am an imposter with less domain expertise. I am only observing what I see from the outside, which is that I don’t view VAR models as an appropriate choice. I really like the concept of them as they allow for interest rates to affect inflation and, in reverse, for inflation to affect interest rates. However, they are designed to help reason about a dynamic relationship between GDP, inflation, interest rates and employment, heavily relying on the huge assumption that interest rates really do affect inflation. Just because we can reason about a system, it doesn’t make the reasoning correct or mean that policy decision making is having the intended effect.
This is why I felt like I needed to step back and correlate interest rates with inflation outcomes i.e. for each consecutive financial quarter after an inflation hike, do we generally see the desired outcome of inflation coming down in the historical data? Is this statistically and practically significant? I don’t think VAR models allow for this reasoning.
Results & Discussion
What does the Inflation Rate and Interest Rate Look Like in the UK?
Interest rates and inflation appear to move approximately in tandem and inflation can rise or fall dramatically in just 1 financial quarter! However, the pink line and elephant in the room is the period following the 2008 financial crisis showing that inflation can do its owns thing.
Correlating Interest Rates and Inflation (Taking A Lag Period in to Account)
When plotting the data we can visually see other factors beyond interest rates that are much more important when it comes to major hikes or drops in inflation
Since a change in interest rates could take time to come in to effect, I used the yearly change in interest rates (adding a lag period in order for it to take effect) and correlated this with UK inflation rates for each consecutive financial quarter.
Here is an example:
In the graph above, 1 year of interest rate changes has passed, such as a 3% rise in the last year, and we have waited a further 2 financial quarters to pass in order to see if it generally affects the quarterly inflation rate. We can see a positive correlation.
Below is another plot but this time after 2 years has passed. Even though the inflation trend is going down, there are more extreme data points scattered either side of the line - the line of best fit fails to explain these major changes to inflation:
If you break up the data and measure the correlation for each consecutive financial quarter after a yearly interest rate hike, you get the correlation curve below. After waiting for 12 financial quarters, the correlation is at its trough:
The graph above has a remarkably similar shape like to the one IMF present in their paper:
However, there are issues that I outline below. Results for correlations are presented in the table below with lag periods up to 24 quarters after yearly changes to interest rates. I have presented correlation values, p-values and R² values as a supplementary (and controversial) measure of model quality:
Remember that we are looking at outcomes of inflation after interest rate hikes for each consecutive financial quarter, so if we are to take correlations as causal, then this suggests that the inflation outcome for a time lag of 1 and 2 quarters stokes inflation higher, indicated by regression coefficients above 0 and low p-values.
If we look at a further 6 to 14 quarters later, we can observe intermittent periods with negative correlations and acceptable p-values i.e. inflation coming down with associated interest rate hikes. However, there are periods where there are no decent p-values, such as 7, 10 and 12 financial quarters after. This signals that this model isn’t consistent over time. I think that in order to take interest rate changes seriously in the linear models that economists use, it should be consistent. However, we know that the world is non-linear and world events can dwarf any modelled effect.
The quality of the model isn’t great but if we take this model at face value and estimate the inflation outcome of a yearly interest rate hike of 3% running its course to 18 financial quarters after, we get a roughly 1% drop in the inflation rate.
I have also included controversial R² values here, which quantify what we can see in the plots i.e. that interest rate hikes don’t explain a lot when it comes to major changes to inflation.
What do the VAR models say?
I have applied a basic VAR model, which can be found appended to my code but as I mentioned earlier, I decided against the use of a VAR model.
Economists talk about supply driven vs demand driven inflation events — can a pattern be teased out by separating these events?
Climate change, wars and technological advancements all affect inflation, so I tried to assign these to each period but realised that it was both an enormous effort and very subjective.
What about interest rates for the US and the Eurozone?
For these economic zones, the results were still not convincing to me.
Is there a rationale to explain how inflation would increase after an interest rate hike?
If businesses face higher interest rates on their loan repayments then I think that they simply pass the costs on to consumer in anticipation. It is much easier than implementing business efficiencies, which can take years. Unsurprisingly, many employees succeed in getting a pay rise in return, as we have seen over the last year in the UK.
BoE’s Interest rate target is 2%. Why?
Why not 5%? What not a range like 2% to 5%? I haven’t seen a decent explanation here. It sounds like a nonsense policy to me.
Closing Thoughts
Central banks are important, despite what the crypto bros tell me everyday, and I understand that central banks feel the need to be seen doing something about inflation. However, I am not convinced that using interest rates in order to tackle inflation works. I am firmly in the indeterminate side of the fence, leaning towards ineffective.
I read from the former Fed head, Bill Dudley, that interest rates are different now that we are now in a new age of an ‘excess reserves regime’. In my view, more data needs to roll in before asserting something like that.
Improvements to my work and the Economics Field
- I don’t like using the CPI in order to measure inflation — we should have a much more far fetching data gathering procedure for it. The ROI on that could be huge. The CPI doesn’t include train fares (thank you for that suggestion Will Page) and given that the micro makes up the macro, this is a big deal when we think about how our daily work habits change over time
- I think a robust analysis of the extreme changes in inflation might improve my work i.e. how sensitive to change are the models. However, given that extreme changes appear to be king, I am not sure this matters
- Fact: the world and the economy is highly non-linear. Non-linear models would be more appropriate than the linear regression models that I have built
- I feel that an engineering approach would help the fields of Economics i.e. redeploy efforts on empirical data gathering at scale over theoretical modelling and the endless theoretical ‘proofs’; an interdisciplinary approach that includes science, psychology, data engineering, finance and members of the public that have actually worked in a business — so not Andrew Bailey)
When looking at the impending doom of my mortgage rate hikes, I found my analysis upsetting — I hope that somewhere in my code I made a big mistake and that I am wrong about interest rates