Inflation forecasts or irrational exuberance?

If your eyes glaze over at the mention of inflation, take a look at this and feel free to go:

My own calculations using forecast figures from the SMP and inflation data from the ABS

Forecasting inflation

Forecasting is an unforgiving art. There are an infinite number of wrong answers, and one right one. The years since the GFC have been particularly unforgiving for central bank forecasts, and no forecast has struggled so much as inflation. First too high, now too low, inflation has a mind of its own which central banks have yet to fully discern.

No surprise then that central banks have grown more cautious. They have commissioned research into their forecasting track record to better understand the scope of the problem, and what might be done about it (RBA, RBNZ, and BoE). Forecasts are now accompanied by caveats, confidence intervals, and cautious language.

An example of representing uncertainty better (Reserve Bank of Australia Statement on Monetary Policy Feb 2015)

While helpful, there are still two small issues:

The reports I’ve read primarily assess forecasts for accuracy. This involves calculating the difference between the forecast and reality and adjusting it to remove the sign.* This shows the error’s magnitude – the higher the number, the greater the deviation – but not the direction of the forecasts or actual inflation. Some reports do also test for bias – the degree to which results repeatedly skew in one direction – but not all.

Second, most of the reports are from before 2015, around the time when problems with inflation started to appear.

I want to see recent results, so I’ve started building a data set from publicly available forecasts. I will eventually do this for all the major central banks, but started with Australia because it has enjoyed uninterrupted growth for the period when the rest of the world has been in and out of crisis. This should minimise forecasting errors from big shocks like Trump, Brexit, or the Euro-crisis.

Persistent errors in the same direction

Inflation forecasts between 2005 and 2014 missed sudden changes in the inflation rate, like in 2006, 2011 or 2012. Some of these mistakes were over-estimations, others were under-estimations. The errors are not surprising given the global recession, and lots of fiscal and monetary stimulus going around.

My own calculations using the figures provided by the RBA
My own calculations using the figures provided by the RBA

2015 is where it gets interesting. Inflation forecasts have been persistently panglossian over the last five years; forecasts keep reaching for the sky while inflation trundles along below. From 2017 the forecasts sober up a little, but the errors are still all in the same (hopeful) direction.

My own calculations using forecast figures from the SMP and inflation data from the ABS

When the forecast errors trend in the same direction, one of the relationships in the model may be misrepresented.*** This kind of diagram illustrates the issue in a way that accuracy bar charts do not. It also suggests that forecasters may not have updated the model.

Forecasting macroeconomic variables may be the only form of divination that requires a suit, but forecasts can still be useful. We feel more comfortable about an uncertain future if we can attach numbers and neat lines to it. This psychological comfort helps us make decisions about the future, whether to buy a house, or invest in a plant that manufactures fidget spinners. If these beliefs are shared by enough people, they can become self-sustaining, and even ‘true’ by simple fact that everyone now believes them. Forecasts can also give a reliable sense of direction,

If the past has any bearing on the future, forecasts can also calibrate models for economic or social behavior. In practice, these models can often give us a reliable sense of direction, if not complete certainty. Here, mistakes are useful for tuning the machine, but not all mistakes are made equal. One-off shocks, like the GFC, add little to models that, by design, cannot predict the unpredictable. Recurrent forecasting errors are another matter. They may be a signal. Taken in this way, the errors are not so much the inevitable failures that go with any attempt to peer past the veil of the present, but indicators of a more persistent change underway.

Or maybe not. It’s the future after all.

As always, if you enjoyed this, please consider sharing or subscribing!

*Called a root-mean-square-error (RMSE)

** Central banks use a larger battery of statistical tools than the simple RMSE

***The Philips Curve is flattening


YCCATRBA almost sounds like a cool new hip hop track no?

But it’s even better: Yield Curve Control at the Reserve Bank of Australia


At the Reserve Bank of Australia’s (RBA) December 1st meeting they voted to keep all policy options unchanged. This includes its target of maintaining 0.1% on three year Australian government bonds, a policy also known as Yield Curve Control (YCC).

YCC was first announced in March, with the RBA committing to buy any quantity of Federal Government and semi-government (e.g. state government) securities required to reach its yield target. The target was set to 0.25 in March, and lowered to 0.1 in November.

One of the concerns about YCC is that the commitment is open-ended. If investors got skittish and started selling bonds, causing yields to rise, the central bank would have to step in and purchase potentially enormous quantities of bonds.

The alternative approach is the quantitative easing practiced overseas, where the central bank commits to buying a certain quantity of bonds, but does not commit itself to achieving a set target for yields. This limits the central bank’s commitment.

In both cases the goal is to shift portfolios to other assets and reduce borrowing costs. The question is which requires the central bank to do less?

RBA Government Security Purchases
Purchase DateQuantity
(Billions AUD)
Total (Billions AUD)Purchases as a % of outstanding long-term government securities
October0.063.0No data on outstanding securities
November5.068.0No data on outstanding securities
Data drawn from the RBA outstanding debt securities database and purchases announced in monthly monetary policy decision press release.

Still, this quick and dirty calculation suggests the RBA has been able to anchor long-term rates quite quickly at minimal cost.** This is good news for advocates of YCC, especially when you consider that QE has left the Fed and the Bank of Japan with bond holdings in excess of 30% or 40% of GDP.

Lets quickly take a look at this from the Government’s perspective (drawing on the Australian Office of Financial Management’s data). We see an increase in long-term security issuance from April, just after YCC began. At the same time, short-term security issuance starts to decline from May. In other words, the government is taking advantage of the RBA’s effort to keep longer term borrowing costs down and loading up on longer-term debt.

Interestingly, this surge in issuance does not seem to be putting pressure on the RBA’s peg. Between May and July, the government raised nearly $100 billion in long-term debt. I do not have data showing breakdowns by maturity, but presumably some quantity of that was 3-year debt. Despite the increase in issuance over the period, the RBA did not need to purchase more bonds to maintain its peg.

There is clearly plenty of demand for government debt right now. The risk is that if private demand for public debt at that price were to decline (for whatever reason), then the RBA would need to purchase ever larger quantities of the debt to maintain the peg. For now, that does not look like a concern.

As I’ve said before, the government should take advantage of this to borrow for an ambitious investment program. I remain skeptical of stimulus overly reliant on tax write offs and incentives.

The RBA separately announced a more traditional $100 billion QE program last month, with 19 billion already purchased. I will revisit these figures once more data on debt issuance becomes available.

*A small purchase was made in May but it did not change the overall quantity, so was probably below $1 billion in size. June and July had no purchases.

**The calculation is not perfect because I don’t have data on how exactly the RBA is spreading its purchases along the yield curve, and whether they are also purchasing short-term treasuries.