Transcript of Question & Answer Session The Why, How and What of Forecasting

Thank you, Marion. We are going to move on to the Q&A now. We do have the Pigeonhole app and the details are on a QR code on your table. But we also have roving microphones, and I’m really encouraging people today to actually – if you’re comfortable doing so, please stand up and ask your question with a mic. If you just pop your hand up, one of our theatre staff will find you and come to you. If you do ask a question with a mic, please keep it pretty brief; and questions and not statements, please; and please do let us know who you are and where you’re from. We do have some media in the room today, and Marion is happy to take questions from media; but, again, please ask them through the microphones and identify where you’re from. I’m going to start with an opening question, but then I will throw it to the floor.

Marion, you talked a little bit about the kind of breadth of data and information that you use, and one of those elements is the Business Liaison Program, which was a program I was lucky enough to be a part of a few years ago when I was at the RBA, and I know we’ve got some members of the team here. Could you tell us a little bit more about how that works, what sort of information you get from it and what are some of the kind of key insights that you’ve got over the years from liaison that you might not have got elsewhere.

We have built up the liaison program in early 2000 – I think it’s 2001, if I remember; Virginia might be able to correct me on that one – and we have offices across the country in the major states. We’ve got one here in Perth; we have one in Queensland, Melbourne and in South Australia, in Adelaide, as well, and there are small teams that are there. We also have a team in Sydney that covers the Sydney area for liaison. And the purpose of this program is really for us to get insights from businesses, in particular on the ground, of what is happening. So we’re covering different sectors and we’re covering different themes for us and for the Board to get the best insights possible on this exercise that is called ‘now casting’: that it’s really important that you know what is here and now. That is one part we get from them. We get a lot of colour on what the underlying drivers are. When you have to model something and when you have to pick a model, it’s actually really important to know what is driving a relationship, so those insights are invaluable. Businesses also can give us a bit of an insight to how they’re looking – thinking about the future, and that also informs judgement we put in our forecasts. So we get a lot of that colour and qualitative measure from our liaisons. And it’s a structured program. We’re often going back to the same contacts, but we are also collating that information in our qualitative results. So we get what we call an ‘index’. We’re putting that together. It’s a Likert scale for those who are – I think, anyone who’s been through our representative offices knows it quite well. But it is a technique to pull that information together, in a kind of aggregate sense, to give us a sense where things are tracking.

And I’m very pleased to be here, and I was actually here in November – the first time after lockdown – to, in fact, visit our Western Australian office. And Aaron’s here and Stephanie’s somewhere; Stephanie is over there. Not only do they kindly host me, but actually they set up programs for me to talk to the businesses here as well and hear what they have to say; and they’re usually also interested in what I have to say, which is nice too; and that’s an integral part for what we’re doing.

What are some of the insights? We’re here in Western Australia; I cannot be here without mentioning mining. So, obviously, you all know there was a huge mining boom that went on about a decade ago, give and take – it took quite a long time – and actually, for us, it was really important to understand what the investment profile that was involved with that. And I remember – I think Virginia was heading the office at the time; she kept a spreadsheet with projects. Because they were so huge and big, they were affecting the aggregate outlook for the economy, and actually it was liaison critically that allowed us to do that. I don’t think, without liaison, we would have had access to those data. So that is probably one good example among many, many, many I could recount where we got information about a critical development that had impact on the aggregate but where you just can’t get the kind of standard ABS data, because it’s such a specific, unique event.

Fantastic. I’m sure there are some of you in the room who have been part of the liaison program. But, if you’re not and you want to get involved, I’m sure Aaron here would be happy to have a chat with you. Do we have any questions from the floor? We have a couple of mics. No; all good.

I want to talk a little bit about the accuracy of forecasting and you put up the confidence intervals that you have, and they’re pretty wide ranges. I think to people that have been involved in economic forecasting, that’s not necessarily a surprise, but it might be a surprise to the broader community about how difficult it is to actually forecast even a couple of years into the future. How does the RBA kind of rate compare to other central banks or other forecasters in that regard? Have we pretty much similar kind of confidence interval, similar kind of kind of accuracy, or are there things that we should be doing to keep improving that?

I would make the bold claim that I think forecasting is a difficult business for all the central banks. No one saw the recent inflation surge. At least, if you go back, certainly no Central Bank called it out either and that is not a surprise. So some of these areas, as I’ve talked through, don’t actually come from forecast mistakes, as in the sense that you make an intellectual mistake in your forecasting, but there’s just shocks that arrive. It’s very hard to predict a pandemic; and, even if you are someone who is a health scientist who said a pandemic would come, you’d also need to tell me exactly when it comes for me to put that into a forecast, or it’s the same thing with global development. The war in Ukraine: that had a really big impact globally. Again someone who knows a lot about geopolitics might have said, ‘Oh, well, I could have told you that something was brewing.’ Well, for me to put in a forecast, I actually need to have the date for that as well and the extent as well. And so those exogenous shocks happen to all economies and no forecaster can overcome those in a systematic way.

Are there things we can improve? Absolutely, and we have published the annual review I talked about that goes to the Board. We have published that – I think it was in November; Aaron, do you remember that – we’ve published it in the November SMP; I think it was either the November or February SMP. We actually published that. We sometimes publish it when we think it’s particularly interesting and it’s a part in the statement on monetary policy. And one of the big lessons of the last two years has been that our models just haven’t been geared to deal with supply shocks because we haven’t had them for such a long time. You, of course, pick a model that kind of has done really well in explaining the recent history, and that’s the nature when your economy changes. Those models may not perform well anymore, and so one of the big learnings has been that, well, maybe supply shocks are going to be more frequent. I’m not quite sure I’m predicting another pandemic, but actually the other nature of supply shocks are coming from the energy side. You can see that there’s a link, and the Governor gave a very interesting speech – for those who want to see that – in November, I think it was, on that topic on ‘do we think we’re in for a future of more supply shocks?’ And, of course, if you think that, there’s climate change – and there’s a lot of evidence for that – you will have more severe weather events, and one of the supply shocks we had last year were the floods and they were significant for inflation. So, if you think those are becoming more frequent, you have to gear yourself up to developing more to sort of contain the supply shock, and that’s kind of the task for us going forward. We’ve dusted off some models that do a little bit better, but we’re actually also actively developing models that actually allow us to handle supply shocks better in our overall forecasting framework.

It’s a big task and, if I talk to my colleagues in other central banks, they’re all facing that very same problem at the moment. So, there’s no magic thing I can take off the shelf, but it’s just what we do as forecasters that we have continuous development. But that would be one I’m pointing out where I definitely think there is work for us to do.

Thanks, Marion, Paula Gadsby from EY. I just have a question around services inflation. So there seem to be upside risks with services inflation, and what we’ve seen in a lot of developed countries overseas is it’s been quite sticky to bring down. How worried is the RBA about this, especially in regard to your forecasts of getting inflation back to the target band by mid-2025?

Thank you very much for that question. I think, if you heard the Governor’s speech yesterday, he did mention that that has been a consideration. And, in fact, you are right: that we have seen the services, the domestically generated component, in other countries being a bit more persistent. And talking about ‘bring it back to the lessons we’re learning’, about a year ago, when in other countries inflation picked up earlier than here – we opened later – the big question was, ‘Well, are we different, or are we just later?’ The answer was: probably just later. And I think one of the lessons was actually, well, maybe we should have given a little bit more credence to that story as opposed to thinking, ‘Well, here’s a whole lot of reasons of why we’re different.’ So, by the very same token, you could look now at that persistence of services inflation in other countries and say, ‘Well, could happen here as well,’ so it is definitely a significant risk.

I think to the extent that we can model it and we have embedded some of that in the forecasts; services inflation does come down slower. You saw that supply/demand graph I had, and a big part of that demand persistence is coming through that domestically generated inflation; that shows typically up bigger in the services sector. So that’s certainly true. We will continue watching it though because it could play out either way. But it’s certainly a risk and, yes, we should look at that.

I do point out that some of the other dynamics in the overseas economies are different; they tend to have higher wage growth than here. But, again, last year, a year ago, we thought out all of those not excuses but all of those explanations and thought, ‘Well, it could be that we’re different; it didn’t turn out.’ So maybe we’re kind of having a bit more of an open mind this time around.

Thank you, Marion, Virginia Christie from the Economic Regulation Authority. I appreciate your going through and explaining how complex it is to forecast, and we’re aware of all the challenges in doing that. Looking at the current set of forecasts though and coming back to the issue around inflation being back within the band of two to three per cent by 2025, what’s the likelihood of stagflation being experienced over that period?

Let me answer that in terms of the error bands. They clearly tell you it could happen, but the forecasts also tell you that that is not our central forecast at this point. That is a risk in the current constellation. The Governor has spoken about the narrow path. He has said last night he believes that we’re still on that narrow path. And that narrow path is one where you bring inflation down, but you bring it by maintaining the maximum benefit of employment possible. So, I can’t put a number on you, but it is within those large error bands I’ve shown you, it is an outcome that is possible.

John Poulsen, CEDA Life Trustee. As we all know, monetary policy is quite a blunt instrument and has the lags talked about and also has the uneven impact on people living in the economy. What other tools? Are there any new tools emerging that might have slightly sharper scalpels to them, or is that the only tool?

If you have suggestions, we’re really happy to hear. The one tool people typically look at is fiscal policy. It can be used as demand management. I do want to refer you to testimony by the Governor – recent testimony – where he was asked that question, and what he said is, ‘Look, typically, monetary policy is, in any country, thought of as being a nimble tool to use for demand management, but there is no question that there is an interaction when it comes to demand.’ I’m not going to comment on fiscal policy here. We’re a separate entity and we have a mandate; and our mandate is inflation and, to the extent that is demand driven, that is what our tool is actually built to do.

Thanks, Marion, for coming and also for stepping in at the last minute, or we wouldn’t be able to be here just now. My name is Alison Preston from University of Western Australia. Can you just talk to us a little bit more about what you’ve been learning about the labour market over the last few years and through your forecasting as well; is it behaving the way you would have thought it, and what is it that’s informing your judgement?

Yes, the labour market over the last few years. So the answer is to say, ‘Is it behaving like you’d expect it?’ I can’t really answer that question because I’ve never seen a pandemic in the data or otherwise, so it’s very hard for me to say what my expectation would have been. But we, of course, look at it and we’re learning ex-post what has happened, and I think these are important learnings because they can also inform us of what might happen in future again.

One of the features in Australia—and the labour market pointed it out, in terms of what learnings we’ve had – is actually we’ve been one of the countries where, through the pandemic, labour supply has increased, participation rates have increased. There’s a couple of countries, and the US and the UK I could name there, where it’s actually gone the other way. And I think it has helped – by and large, it has helped us, right? At a point where you have tight labour markets, to have people drawn into the labour force through the participation rate has actually helped us keep some of those pressures a bit lower than in other countries. So that would be one possible learning you can do.

I think it’s still very early for us to draw conclusions and you asked the question, after pondering a future speech, once we have the full set of data to look exactly on what we can learn from this extraordinarily pandemic period in terms of labour markets. What is certainly true is that the unemployment rate has been as low as it’s been in 50 years and so has been a whole lot of underemployment, and I don’t think anyone would have predicted that, if you had told them in 2019, ‘A pandemic is coming and that is where you come out of it three years, three years later.’ But I think it would have been a hard ask to predict any such outcome.

I might ask a follow-up question on the labour markets. Given the extremely kind of low rate of unemployment, have you been surprised by the slow pace of wages growth, and that’s really taking quite a while to kind of pick up there?

I wouldn’t say we have. Over the past year, by and large, if you take out – as I said on consumption, if you take out the interest rate increases, if you take out the housing prices and other kind of the population growth as well, actually, it’s kind of turned out broadly as we had predicted or as we had expected – maybe not as we have ‘predicted’, since I’m not doing the predictions; that’s certainly true. It’s a really hard question to answer, and that goes really to the question of where do you think is that point in the economy where wages and inflation start accelerating or not.

Many, many papers have been written, including internally. I would say there is no single point I can point to, but there’s also a whole lot of literature that actually says this could be non-linear; so the further you get away from it, the more your wages might behave differently. So, to some extent, we’re also constantly observing and monitoring how this is developing and whether it is developing as we thought or not and have an open mind, and it is that continuous improvement where you see wages growth going.

Jordan Murray from The West Australian. Thank you for speaking today. I think here, in Western Australia, we like to think that we carry the national economy on our back. Based on the forecasts that are being put together at the moment, I was wondering if you could speak about the role that Western Australia is having in national forecasting and then, specifically, what Western Australian policymakers can learn from the national forecasts.

This is a really hard question. So, we’re in the business in forecasting to forecast aggregates. We have a mandate for aggregate inflation and aggregate outcomes. And so, a lot of our forecasting work is being really guided by what it means for the aggregate. That doesn’t mean we’re not looking at Western Australia, and I think our liaison program is kind of the key way how we gather that information, if that helps.

What I can say is that I’ve been here during the mining boom, where it felt like Western Australia was developing at quite a – it had very different themes to some of the rest of the country. But what I’ve actually found in my recent visits here is that despite the differences, there are actually a lot of common themes coming out of the pandemic in Western Australia as well, like in the other states. That’s probably how I would cast it, without making this about one state carrying the other states; that’s just not how we think about that.

Thank you. I’ll take a question from Pigeonhole, which I think is quite an interesting one. How is the RBA thinking about the use of advanced analytics and AI in the forecasting process; is that something you’re looking into?

That’s a really – if Luci had been here, she’d get really, really excited about that question, so I’ll try to channel. So, let’s take it one at a time. So, we have heavily invested in data analytics, we’ve been building up internal expertise, and it’s allowed us to do a couple of things that we wouldn’t have been able to do a few years ago, particularly using the large datasets that are available at the moment. We kind of tend to publish them sometimes. They’re sometimes in the statements; sometimes we have a Bulletin article going at that that we’re dabbling in in – you go to Google and you kind of grab headlines, or you’re doing text analyses, and you can see that there’s academic papers out there as well and just see whether we can learn something and whether this gives us some additional insights or corroboration of other things.

So that is certainly an area that we’re looking at. It’s not the core of what we’re doing. And the reason it’s not the core of what we’re doing is, as I explained through here, the forecasts aren’t just about coming up with a number that may or may not be good, because you know you’re never going to hit it. What is really important is the narrative here, developing about that. So understanding what is driving it and having a logic behind the behaviour is really, really important for our job, and I’m not sure whether AI is quite helping there.

But I have an open mind. If someone has an AI tool and wants to put it to the race with our forecasts, we have a very open mind to go into that sphere as well. But, at the moment, we’re not using it for that reason that we actually really think the narrative is important because we have to think about risks as well as the central forecasts and have to be able to quantify them and understand them and understand why we’re tracking along a forecast or why we’re not tracking along a forecast in order to make judgements whether policy needs to be adjusted or not. So that is a much more complex thing, but I have an open mind whether AI can help us with that.

Fantastic. Watch this space, it seems. I want to touch back on the transmission mechanism. And you talked a little bit in your speech about the time it takes for policy decisions to really flow through to different aspects of economic activity, and I think that’s something that’s not particularly well understood, and how long some of these things take and why that is. I wonder if you could just tell us in a little bit more detail about what’s the kind of process there and why is it that things kind of take so long to flow through.

At a very fundamental level, it is because what we’re observing at the end is the result of behaviour, of behaviour of firms and households, and so you put an impulse in – that is different interest rates – and you actually need to give it time for that behaviour to adjust. That is at an abstract level why you have lags in there. And then behaviour of different people, parts of the economy, will adjust earlier or will adjust later and, if you play that through, you’re kind of getting the whole chain of events.

But at a very fundamental level, economics is about the behaviour of people – about understanding the behaviour of people – and, when you’re a policymaker with a tool, you basically want to influence that behaviour to achieve certain outcomes. It adds up to the aggregate; it looks like the CPI. But, of course, the CPI is basically the added-up effect of a whole lot of businesses making decisions about prices, and that’s kind of why it takes a long time to flow through. Different people will react differently earlier, later, differently. A saver will react differently to a borrower, just at the front end of the cash flow channel. That alone, you can see that there are just lags building up that you do.

So our models are trying to capture these at an aggregate level. But, if you think about it that way, you can understand why the error bands any model has are actually pretty large the estimated error bands, not the broader asset. It’s just the nature of the thing we’re dealing with.

A question has come through on Pigeonhole and you did refer to this a little bit in your speech. We had a lot of people on fixed-rate mortgages, which are likely to come off in the fairly near future. Is that something you’ve incorporated into the forecasts: the impact of that?

We have spoken a lot about that in public and in our documents, and there’s different aspects to look at it, so I’m just going to focus – the reason – I’m just processing, trying to think through which aspect I should do. So I’ll just take the economic effect on how that impacts. If you think about it, you have someone with a variable-rate mortgage where the interest rate adjusts much earlier and they face the lower mortgage. Someone with a fixed-rate mortgage will face that same adjustment, just a little bit later. So you can take that into account, but the fixed rate mortgages actually aren’t in the scheme of things that long. I think for Australia they’re long; they’re two to three years. I go internationally and speak to my colleagues from other central banks, and they say, ‘What, they’re not 30 years?’ right, because there are actually countries that have much longer runs on the mortgages. And so, if you look at it that way, they’re actually rolling up relatively soon over the horizon. So, to the extent that we can adjust for that, we have done it, but we’re actually not modelling the pass-through at that detailed level, but – yes. So there is an element by which we’re adjusting a little bit for that.

But it also affects where you just one channel of monetary policy, which is the cash flow channel. As I said in the speech, it is very visible because it’s one of the fastest channels. But there are a whole lot of other channels, such as through housing markets; the exchange-rate channel; and other channels, like consumption and saving channels. They take longer, but that doesn’t mean that they’re smaller than the cash flow channel. And so, in some sense, when you add them all up, that is actually for the specific question of the transmission of monetary policy to employment and inflation, that is actually a small movement where we have it. But it is a very visible one and it is a very important one for those households that are affected by that.

Fantastic. We’ve got time for one last question. Does anyone want to take a last chance from the floor? No? We’re all being very quiet today. Well, I will ask my question then; it’s quite a nerdy one. We finally got a monthly CPI indicator, which has been a long time coming. Has that assisted the forecasting process, and are there other kind of key data gaps that you’d really love to see to help you improve forecasting going forward?

I’ll take the first but maybe not the second one; I have those conversations in private (inaudible). Yes, it’s been a long time. We discussed that and I think it’s just a big undertaking to do that. Has it helped us? I think it has. I think it is early days. Yet it was very timely to get that. I think, when you have inflation moving fast – in either direction, for that matter – getting a more frequent update I think is really helpful.

Because it is early, I think the Bureau of Statistics is still working underneath and refining and collecting more data and updating. We’ve had a box on what it looks like in the November SMP and, when you looked there, you would have seen that the way it’s done is that not all goods are updated every month. And so that introduces a little bit of a problem that you need to assess what information you get every month, and I think some people who looked at it at the beginning have found that it’s a bit more volatile than is the quarterly. And it’s probably not surprising

aggregation tends to kind smooth out errors.

But we do actually learn a lot, and what I have found so far is, if anything, we’re learning more about the components. So there are certain components that we’re actually seeing early. And so looking underneath the hood has actually allowed us to understand and get an earlier look into how some of the components are developing earlier than if we got the quarterly CPI. But, with every new statistic, you actually have to learn how to deal with it, how to get the best information out of it. And I think we’re still in that early period, where we’re trying to see how we’re best reading that new statistic to get it out. But it’s definitely been, already, a really great asset to have.