IMF Holds 14th Statistical Forum on Responsibly Embracing Big Data and AI Opportunities

Live News Network
1.4K followers
Nov 19, 2026
The IMF (International Monetary Fund) holds its 14th Statistical Forum on Nov. 18–19, including a one-on-one discussion on Nov. 19 on how the global statistical system can responsibly embrace the opportunities involved with big data and AI.
Bert Kroese, chief statistician, data officer, and director of the Statistics Department at the IMF, will moderate the event. Christopher Phelan, chair of the Council of Economic Advisers (USA), and Min Zhu, former IMF Deputy Managing Director, lead the discussion to explore topics on trust, innovation, official statistics, and the future of economic measurement.
AI Generated·May contain errors
Okay, we will start the last session.
of this statistical forum.
And I'm really looking forward to this final session,
But I'm also a little sorry that it's over in one hour.
Because we learned so much, we had such great discussions.
And over the past two days we've explored how
Statisticians can use big data in AI to generate new data.
expand the toolkit of macroeconomic statistics,
more timely data, more granular data
overwhelming evidence to observe the changes that are happening around us.
We also discussed how we need to ensure
Our current instruments, such as GDP, remain relevant.
The last session was about how you perform the deflation.
How do you create an AI-themed account?
So we are so eager to do the right thing,
But there's no point in doing the right thing if it isn't put to use.
So we need to talk to our policymakers—what do they need?
What information would be helpful to them?
What are the answers they need to have?
their understanding of the economy of the right foot,
And what does it mean to design targets?
and implement, as well as evaluate, the policy.
Because more information does not automatically mean better analysis.
Economists need to understand what data we put together.
You need to tell us what to do and where their interests lie.
And we need to explain the limitations of the data.
So all the efforts that we make truly lead to
for better analysis, insights, and policies.
We are looking for a continuous partnership.
And I think this forum contributed to that.
So, in this final panel, we will explore these questions,
and I'm really happy to introduce two economists who bring perspective
from research, policymaking, and international finance.
To my right, Brian Wheaton,
He is the Chief Economist at the Council of Economic Advisers.
on leave from his position as assistant professor of global economics
and management from UCLA Anderson School of Management.
Prior to this, he was a Hoover Fellow.
at the Hoover Institution at Stanford University.
His research covers a broad range.
on topics within macroeconomics and political economy,
unified by a focus on the unintended
or unexpected consequences of government policies and other shocks.
He holds a PhD and a Master's degree from Harvard.
and a Bachelor's degree from the University of California, Berkeley.
So, welcome, Brian.
Thank you.
Min Zhu, I think many people still know you.
You worked at the IMF before.
Texting
You are now a member of the Chief
Executive Advisory Council in Hong Kong
and once again, former Deputy Managing Director,
and I just learned that you instigated
a.v.
the first statistical forum at the IMF when you were here.
Nicolai
So, thank you.
It was a long time ago.
We are still honoring your inheritance.
Previously, you were the Deputy Governor of the People's Bank of China.
Winther
Executive Vice President of the Bank of China
and your experience across national policymaking,
international institutions and finance
brings a global perspective
on how better evidence can support economic decisions
across countries at different stages of development.
So, we'll dive into the questions, and I'll start with Brian.
Brian, maybe you can first explain to us,
What does the Council of Economic Advisers do?
Sure, happy to do so.
So you can think of the Council of Economic Advisors, or CEA for short,
as something akin to an internal consulting firm for the White House,
in the sense that when individuals and other components of the White House,
or some agencies around D.C.,
are considering a policy, they may come to CEA and say,
hey, can you tell us what the likely
What are the impacts of this policy to be on?
Usually GDP, inflation, wages,
Some of the standard outcomes they are interested in.
And there are about 40 of us at CEA,
so 10 senior economists, each specializing in a particular field within economics,
about 10 junior economists, our chair, our vice chair
and myself, the chief economist, plus some other support staff.
Wonderful. So you're very close to policymaking in the US,
So I think the question and your answer are very interesting.
Thank you
Well, for decades, economic policy has relied
on the traditional toolbox of macroeconomic statistics.
Economies undergo rapid technological,
for
demographic and structural change
What information do policymakers most need?
to understand economic performance and make effective decisions?
at
And especially, what information
you
What is currently missing to evaluate those changes?
Yes, it's really a great question.
so
I think the way of answering it
is to begin at the end and work backward.
on
That is to say, what are the questions we truly wish to answer?
What is it that we truly aim to address with the data we collect?
And, you know, broadly speaking,
I think we want to be able to make a statement of
Our people, is productivity improving?
Are people's living standards improving?
What's going on with these, you know,
What are the major issues we care about in the economy?
And, you know, for those purposes,
Traditional macroeconomic statistics have often functioned well.
You know, national income accounting,
collecting data and compiling data on GDP, investment, and inflation
These will continue to remain at the core
of extensive policy-relevant analysis.
But, you know, a major challenge with these data, of course, you know,
when I'm back at UCLA teaching my students about macroeconomics,
when we talk about real GDP, we often describe it as
Well, you know, real GDP is what you get when you calculate the value of everything
that was produced this year and assign the prices from the previous year.
One of the biggest challenges, as I'm sure all of you know,
Well, how do we think about last year's prices?
for some state-of-the-art SI model?
How do we think about the prices from a previous year?
for something that did not exist in the previous year?
And, you know, statistical agencies come up
with methodologies for addressing these challenges.
But there isn't really a silver bullet,
some permanent solution that we can adopt and say,
Hey, we don't have to think about this anymore.
This is what we're going to do going forward.
It will be compatible with any possible
technological innovation that is in the future.
So it really does require, you know,
a lot of continued introspection and continued...
debate among economic statisticians, people like you
That's very important, you know, to give
You know, for example, Si has created
a lot of challenges surrounding measurement, you know
Like I said, we didn't have some
just a few years ago, among these cutting-edge models
The BEA just recently, about a month and a half ago at this point,
updated some of their methodologies regarding how they measured
the price level for software products
So it is encouraging that agencies continue to evaluate,
Re-evaluate, think about these things.
But again, there hasn't been a solution that is a silver bullet yet.
Continued thought, continued introspection, I believe, are necessary.
One thing I do want to emphasize, you know,
another key role of collecting these data,
these macroeconomic data are used to analyze the effects
of government policies that have been put into place.
Unlike decisions made in the free marketplace, there isn't some sort of
built-in self-correcting mechanism in all cases for government policy
That is to say, if a decision is made
in the private marketplace to make a certain investment,
and it turns out that was an inefficient investment,
That company is likely to go under.
If there is more demand for something, then the price system is going to rise.
to function as it does, prices will rise, new entrants will arrive
into the market, and some of that demand will be met.
In the case of government policy, you don't have some
among those self-correcting mechanisms, and that is one of the reasons why
Such data collected by statistical agencies are so valuable.
They allow us to go back.
and retrospectively evaluate the effects of these policies.
But, you know, I would also emphasize that this administration
especially considers it a desirable feature of the market system that,
You know, even if policymakers don't have the absolutely perfect data
At every single juncture, you actually don't need perfect data.
as an economic policymaker in order to enable many
of these functions of the market economy to keep operating.
Now, the...
Better data is something that is very important.
for the reasons I outlined earlier.
Ensuring that productivity is improving and people's living standards are rising
are improving, tracking our overall performance.
And also on a more specific basis,
tracking how specific policies are functioning, tracking how things
are going years after a policy has been implemented retrospectively.
But, you know, the fact that the market system does have some
Of these built-in stabilizers, one is a helpful tool.
We don't need to measure every single thing in the economy,
although it can be helpful for evaluation.
Thank you, Brian. And indeed, statistics
are meant to inform policies before they are made,
but also to evaluate, to learn from it,
And government is different from the marketplace.
Last week I read, or I must say,
I glanced through the president's economic report.
It's more than 400 pages, so I didn't read every single letter.
But you measure information gaps there,
specifically in the sense of supply chains
problems with incomplete information
Is that one area that, apart from the productivity you mentioned,
and it's also something you really
Yes, that's something I was planning to go into detail about.
Nice to do so now.
So I think one of the crucial gaps in terms
of the current data availability that we have experienced
with what we do on the Council of Economic Advisers
is the fact that the input-output tables that we work
are limited to a few hundred sectors.
The input-output tables have
They have been tremendous tools for what we do, tracing.
A finished good backed up to its initial inputs,
gaining an understanding of the policies that affect, for example, downstream sectors
or upstream sectors, and how these effects permeate throughout the economy.
It's an important role that they play.
but the fact that they lack a certain level of granularity
that does exist in some other settings prevents us from...
Making statements at very granular levels, such as identifying
key choke points related to critical minerals in the economy.
That's something this administration has been thinking about a great deal.
Thank you
Identifying key choke points related to pharmaceutical supply chains,
Another important thing.
But at the current level of granularity of input-output tables,
shall
A lot of these things are combined into a single sector.
And therefore it makes it a little bit more challenging to...
answer these questions in a deeply
you
systematic way. So that's one gap
that we would agree.
It's a challenge we gladly take up. We're working on exactly that.
Ha.
It's good to hear that you mentioned it as well.
in the economic report
Thank you
Min,
the same question for you, and
Basically speaking, from China, but maybe
as well as other major economies.
Where do you see the most consequential
shall
information gaps for policy
makers, including gaps related to
The economic impact of AI?
And how do these gaps differ across countries?
Are they the same across the countries in Yorkshire?
you
First of all, thank you for the invitation. It's a genuine honor.
Ha.
And a pleasure to be back and to see so many, so many faces.
Thank you as well for the invitation.
It's really nice to be back at the front, like coming home.
And especially, I thought you picked a very young photo of me.
The picture was taken...
I was talking, I think it was 12, 14 years ago, so thank you very much,
But it really shows how old I am, you see all those white hats.
My LinkedIn is the same.
So you're asking a very good question.
I think the truly missing part, the challenging part,
is really between the fast-changing economy, fast-changing technology,
and the data we have, and the measurements we have.
I think this is really the key challenge, not only for statisticians,
but for all the governments, and for all the economies in the world,
And for everyone, I think it's a major issue.
You just mentioned China; if you think about China,
China being the fastest-growing economy,
and the most profound structural change in the economy,
and also have a digital economy grow extremely rapidly.
This is good news, but it poses a tremendous challenge.
How do you measure the economy?
So in China, I think the experience is really divided into two parts.
The first part is how we use big data.
artificial intelligence to collect data.
I think this is very important.
In the past few years, I have merely observed the Chinese statistics.
Viewers can significantly improve their work.
For example, they create a statistical cloud.
So centralize all the information,
Bring everything from above; everything is online now.
Digital connections are increasing now, and I think it's very important.
And the qualification, review, and all the checking,
All of those things are now done online, and I think this is very helpful.
And also in the agricultural areas, a lot of satellites,
remote, you know, sensor systems, drone systems.
So in the agricultural areas today,
Most of the statistics are generated by AI.
You don't have to go to the field, do the sampling,
You know, and test the harvest. You don't.
I think this represents a significant change in how this technology is used.
To improve your statistical work as well, right?
But there is also, particularly in using the AR area,
and for example, in the population survey,
For a large survey, right?
Usually, when you do sampling, you move among a large number of people.
In China, every five years, you conduct a census.
surveillance, right? You were mobile, millions of people
to do that given we have 1.4 billion people
but now it's all online, it's all
On mobile, equipment seems much easier and accuracy has improved.
dramatically, so I think artificial intelligence is big
Data can really help a lot in statistical areas.
Thank you.
Just my observations in China could improve efficiency,
broad coverage, and also granularities
you
all those classical statistical issues.
for
You can use the technology to help you.
to save a large amount of labor force and improve efficiency.
Watching!
The second part is not easy.
How do you measure a changing economy?
I think that's a major issue.
And in China, obviously, we are undergoing truly profound structural changes.
The labor-intensive industry is in decline,
Housing is on the downturn,
but the tech manufacturers are really moving very fast,
The AI sector is moving fast.
How do you capture those things?
I think this is really, really a big issue.
There are several things.
I think the cost-based approach remains the fundamental principle.
Because you don't want to inflate all those pieces of information.
Because there is a core,
we observe that there is a core to the digital economy,
which is very much the software.
AI and hardware, very narrowly defined.
But there are also people
Who usually uses a broadly defined digital economy?
Which company uses digital technologies, AI, and whatever else?
In China, narrowly defined, it is roughly around 11% of GDP.
but broadly defined, it can be around 55%.
So the first number indicates that it's a genuine AI sector.
The second number indicates the diffusing component of it.
How much of the entire economy is made up by AI technology?
So these are two different numbers, two different things.
We have to understand that these are two very different things.
But how do we capture those things that become an important issue?
Core spaces do the core, because you don't want
to inflate the entire figure.
I would like to emphasize those things once again,
otherwise,
I have seen various cases of people inflating these numbers.
And I think that's very important.
The second challenge is for artificial intelligence,
It cannot place them into the AR sector.
Since the company has an AR title, right?
You can't even set it by sector.
because it really has to go down to the job level.
You have to understand, it's really
People use artificial intelligence in the process.
Then that makes sense, right?
So it's the survey data not only at the industry level,
not only at the company level.
You really need to go to the assembly line.
You need to go to the job position to understand the whole thing.
Granularity is really the key issue.
So you need the entire system to get the entire thing.
So we conducted some sample studies to obtain that information.
What we realized, compared with
The information we obtain from the survey is at the company level.
And at the job level, it can be very different.
I think it is very important to understand these differences.
The third issue arises because of the key issue you are using.
Big data is data.
So what China tries to do is start with data.
China did one thing that was quite advanced, I would say.
because China established the State Data Bureau.
The bureau doesn't do any statistics.
but the bureau responsible for data collection.
Is that the NDA?
Yes, yes, NDR, yes, NDR, yes.
It's part of the data collection.
data production manufacturing
data infrastructure
data valuation
data technology
They don't do any statistics.
but at least they have someone
to provide data to the statisticians
and with certain quality and evaluation.
I think this is extremely important.
And in addition to this data bureau, I think the three key issues,
because ownership is a very important issue,
if you do the SNA, the ownership and the market,
You know, really play a key role in valuation, right?
But who owns the data?
So I think this is very interesting.
China categorized ownership into three distinct categories.
The first is called holding rights.
It's not a complete one you hold.
The second is the right to production,
because you have to generate the data in order to make it usable,
for example, for insurance companies, for biotechnology companies,
For whatever purpose, whoever produces the data can derive value from it.
The third category is trading and operational functions.
If you trade data,
You do those things so you can profit from data.
So, with three definitions, the data is tradable, and you have value.
So then data go beyond cost to the market.
Value, then proceed to your SNA classification.
So then, I won't say we resolve all the data issues,
but it truly provides a kind of bottom line,
a cushion for data quality, data valuation, or whatever else.
I think that would make establishing a statistical bureau much easier.
To understand the data, to evaluate the data,
and to input data into the national accounts.
So I think those things are observed from China.
Thank you, Min.
Indeed, the MBS uses satellite data and digitalization.
We are working together. They are attending this conference, meeting tomorrow.
Thank you
That's great, and you're right,
You distinguish between AI usage and AI production.
Thank you
Looking at its use, I mean,
And that will also be a question for Brian.
for
Many major economies are currently facing the same challenges.
shall
around prices and demand, around trade,
on technological change in AI
I
jobs and skills, housing and financial stability
day.
the transition toward new sources of growth.
you
One way or another, this is what occupies economists.
and policymakers in all countries.
Thank you
Ha.
Brian mentioned productivity.
imagining that as one of the issues, the supply chains.
for
You measured AI adoption usage.
What would be top of mind for policymakers
at
In China, to get more information about?
What would be most important to them?
you
Or perhaps other emerging markets as well?
so
Well, you always want to get more information.
Yes, that's precisely why we still have a job.
med.
But you need clear,
qualified information
I think this is the most important thing.
I think today the most important thing you have to understand is that change begins with awareness.
What is the AI industry?
What is the valuation of the entire activities?
I think that's really,
There are so many different ways to calculate AI valuations.
all those things, the supply chain, and in the AI supply chain as well.
I think that's a major issue.
But I would like to say one thing—it's truly fascinating.
but it is not necessarily linked with the AI industry,
But it's interesting.
Because change is so rapid.
What I realized, the one thing is,
We need information on the evolution of AI, and how much AI is transforming jobs.
on occupations, and understand the impact on education.
Education
I think this is the most interesting. Think about that.
When the school recruits people,
The students will take four years to go through the entire process.
When you graduate from the school, the job description
is completely different from the year you're getting in.
Because AI evolves so rapidly.
I participate in a project we call the China All Star Net.
So it's job market information.
We get approximately 700 million entries from all the advertisements,
from the internet, from the platform, all the data.
What we found is that the occupation title doesn't change very much.
but responsibility.
It changes quite a bit indeed. The scope changes significantly.
So it's not a limited job; it truly expands the job's requirements.
And between these two jobs, it limits something in between.
But the most important thing is that the job changes very quickly.
It's not in ten years that the job definition will change.
The job description will change approximately every 2-3 years.
That means when you graduate from school in four years,
Your topic is quite different.
So I think that information is very important to understand the nature
of technology, understand how much technology impacts the market,
and understand what those impacts
As I mentioned regarding the case in the education sector,
but more broadly on people's well-being, on income distribution.
I think those are real things happening today,
And we are falling behind.
So, the impact of AI on labor, on education,
the impact on the industry's production
Thank you
Brian, you've already mentioned productivity and the evaluation of policies.
Anything to add on this?
Yes, I thought Min mentioned a number of very important ones.
shall
and fix the issues.
You know, one thing we've struggled with as we've been trying
to analyze the impacts of SI on the broader economy
you
Is it really the case that the survey data just isn't available?
to do this the way that we would like to.
Ha.
You know, one of the standard data sets used
for this is the Census BTOS survey,
which basically asks,
Has your enterprise used AI and productivity tools?
But that's not quite the same as,
Is your enterprise using it intensively?
Kill, you know.
There is a measure at the extensive margin,
There is no measure at the intensive margin.
We do not have these measures of transformation.
of the jobs in the way that we would like
in order to be able to compile these reports in a more comprehensive manner.
There are some private sector alternatives.
that have attempted to measure the adoption of SI,
And perhaps we'll discuss some of those points a bit later.
And so we can use some of those,
But they still have some flaws here and there.
So definitely, that, I think, is one of the most promising areas.
in addition to what I mentioned earlier about
input-output tables for additional data availability
Great, thanks Brian. Good that you were on the same page about that.
Then let's move on to policymaking.
Brian, you're in the White House,
So you're very close to economic policymaking.
Texting
If there were better information,
What would that imply for improved policymaking?
And perhaps you could even provide a specific example of a policy decision.
hi
where better and more timely data might have made a material difference
Nicolai
updated information available to policymakers
At the moment, a policy decision had to be made.
Yeah, so another great question.
Thank you
Winther
So a few different points.
One of them returning to the input-output tables.
So a lot of the questions that we analyze,
shall
a lot of the analysis that we do at the CEA,
involves input-output tables
involves observing how shocks propagate up or down supply chains.
you
And in many cases we have to combine,
due to the granularity issues there,
Ha.
We have to rely on the fact that certain sectors are combined.
Consequently, we need to make certain assumptions about, you know,
Each individual subsector within this sector has similar linkages.
to the other subsectors within the sector.
This makes it more difficult to target specific sectoral policies.
and so I think it is certainly the case that if we had access to better,
More granular input-output tables would give us greater confidence.
about policies that must target specific sectors,
of which there have been many in this administration
and recent administrations
to address issues of supply chain vulnerability and choke points
That will become a major phenomenon, especially after COVID.
That's the one set of decisions that could be,
I think it was made more timely.
and have more analysis
More credible analysis, back it up with evidence
before we pull the trigger
Another thing that I think would be a significant benefit,
one of the major goals of the administration in the first year,
was passing the OBVBA, the One Big Beautiful Bill Act.
Um, and
And the key changes in the OBVBA were structured around expensing,
so restoring full expensing of equipment,
Thank you.
introducing temporary full expensing for factory construction.
restoring full expensing for R&D expenses
you
And one thing that we were very interested in
as the policy was being designed
very
exactly how we should structure these incentives
whether the manufacturing structures, incentives,
Much.
The expenses should be in the factories or more broadly.
Thank you.
We are often relying on data that comes with a delay.
of a couple...
quarters, or in some cases for more detailed data sets,
you
lags of five years, if there's something
That relies on the economic census, for example.
for
And so I think this is a general comment,
something that policymakers around the world face, you know,
Watching!
You often have to work with slightly outdated data.
You're often a little behind the eight ball because of that.
Now, some of these private sector data sources, once again, have helped sort
to bridge the gap in some cases and provide us with more up-to-date data.
But those have their own problems, don't they?
Those datasets are often convenient samples of one kind or another.
Something that bridges the gap between both of those,
I think it would help in that dimension.
Very clear. And indeed, official statistics are not as timely.
as the fast indicators from the private sector,
But there are other issues involved. Let's get to that.
Thank you
Min, we were together at this academic symposium in Shanghai.
in June regarding the impact of AI on the Chinese economy,
and you delivered a keynote there on how AI, particularly agentic payments,
shall
can transform the international financial and monetary infrastructure.
There were many presentations there using alternative data sources.
you
I remember there was a presentation.
Ha.
on exactly the topic you just talked about
The impact of AI on the labor market.
They used data from five recruitment websites in China,
Put it together to form a China AI.
Thank you
I forgot the name, but that's what was being done there.
but it made use of big data, of alternative data
for
which has its flaws, because it is incomplete,
at
And how do you explain that to policymakers?
you
Because Brian clearly said we need fast information to make decisions.
so
But then we turn to information such as the information
med.
that was presented at that conference.
So how do we communicate effectively as statisticians to policymakers?
What is the quality of the data, and what can they use it for?
On the one hand, they need fast information,
On the other hand, there are some issues with that data.
So what would your advice be in that situation?
I think this is a great question.
I don't know whether in the advisory council you have
a statistician there
Perhaps you should have a chief statistician.
next to your chief economist
Because today, you engage in any kind of policy-related thinking process,
And you must understand what the data is.
That's the key challenge, isn't it?
Basically, I see that today there are two groups of data.
This is official data, isn't it?
It's a national account, basically.
GDP and all those things you mentioned,
interest rate, growth rate, whatever.
But there is also other data, mostly from the private sector,
The internet and websites, we refer to as big data, right?
So, these two datasets are both useful and very important, right?
But I think for policy, the key issues we have to understand,
We still use classical statistics, the national accounts,
To formulate policy, to review the course, to see how things unfold.
I think this is very important.
And also, it seems to me, we use more big data,
Not looking back to check, sort of.
What is going to happen in the past?
because those data are unstructured and those data
are not necessarily historically consistent with everything, but rather more along the lines of
occur simultaneously at the same time on a very different surface.
But I think this data can help us accomplish things.
to look into the future, to do the forecasting.
So these two different datasets indeed serve two distinct purposes.
I still think classical statistics should conduct the policy review.
to describe what happens in the real economy.
But I think big data can really help us look into the future.
I think those are the two things policymakers need the most.
But between them, policymakers need statisticians.
to explain the difference
And to tell, even in big data, where does this data actually come from?
What is the qualification, what is the classification?
You know, although everyone says it's a black box,
Right, what does this black box mean?
What is the confidence in the areas, what are the probabilities?
I think this is very important because the key issue today is
For many policymakers, the issue is that they receive so many reports.
using and assigning
So many different sources of data without clarifications.
I think that could be very misleading.
In that sense, for policymakers, anything you read, anything you have
You have to understand where those data come from.
You must understand the clarification, you must understand.
the quality, and you must understand where those data come from.
So you need a chief statistician.
I hope you do have statisticians in the White House.
Yeah, so if I could follow up on that,
I think you make a great point.
So we did, until recently, although
His title was not Chief Statistician.
We had a senior economist for statistics and data.
And I think one of the things that I've gotten...
But he is still a senior economist, not a senior statistician.
You know the difference.
You know, because the DNA is very different.
Well, he now heads the BLS, so he is a statistician.
Oh, that's good, yeah.
You know, basically, statisticians are more stubborn, which is good.
It's a good stubbornness.
We are, I can confirm that.
It's a bit stubborn, so DNA is very different, so anyway.
And chief statisticians are even more stubborn than
Normal statisticians, I can certainly hear it.
So, Brian, what does this mean for statistical offices?
I mean, you have the BEA, the Census Bureau, the DLS,
And we have this entire world of data beyond statistics.
You're sitting in the lighthouse talking to the other economists.
What role do you see when looking at all different kinds of data?
for statistical offices in the U.S., but perhaps broader?
I think it's an important role.
You know, at the CEA, we receive the official statistics.
for each big data release, preceding virtually any other agency.
This is, you know, one of the reasons why I got such
an appreciation for the work that our senior economist for data
and statistics was doing while he was there.
So, you know, as an academic economist, you're used to,
You know, you do a little bit of due diligence regarding a data set,
where it came from
I tried to get a sense of how it was collected, but really
understanding the nuts and bolts of every little seasonal adjustment
of what might be causing certain quirks that you didn't expect
in each monthly CPI release or each monthly jobs report.
I truly developed a deep appreciation for both their diligence and,
Yes, sometimes statisticians can be wonderfully stubborn.
But in terms of the role
of official statistical agencies going forward,
Thinking about this as an economist, I believe one argument would be,
Well, you know, if we're in a world of big data and SI,
where it is possible
to carry out far more efficient data collection than before,
is this taking what was previously a very high fixed cost,
a product with extremely low variable costs
and substantially reducing the fixed cost?
And if you believe that, then perhaps the view is, well, actually,
this reduction in barriers to entry means that there is
a relatively larger role and a relatively more significant role
for private sector data collections going forward.
I think that's part of the story.
But I think there is another aspect to the story as well.
So the other part of the story
Is that an important omitted variable there?
Official statistical agencies bring a level of credibility, a level
of institutional significance that has been established over many years.
This cannot be dismissed when it comes to official statistics.
Public trust in the reliability of the statistics is essential.
So what I see going forward is somewhat
of a symbiotic relationship between this growing body of the private sector
or private organization data projects like this.
Billion Prices Project, Nielsen
Retail scanner data, that kind of thing
and on the other hand, official statistical agency publications.
And to the extent that these things move in the same direction,
that they reinforce each other, seem to yield similar results,
I think that will increase public trust in both.
To the extent that they differ, you know, like what we've seen in some
notorious cases that many of us in the room are probably familiar with,
you know, a case of Argentine inflation in the early 2010s,
the fact that a large number of countries' official exchange rates bear
no resemblance to the unofficial black market exchange rates.
To the extent these things diverge,
I think it erodes public trust.
Once public trust is lost, it can be very difficult to restore.
So I do think that an important thing
that most official statistical agencies have to bring to the table
is that, you know,
institutional history and that sense of public trust.
And so, as a result, I find it hard to believe that they're going
to be displaced by private sector data initiatives.
I think there's going to be a symbiotic relationship.
If I may, I think this is a very important issue.
Thank you
I mean, in the real world...
You can see the big data, all those, you know,
for
The information sectors, companies, try to eat your cake.
at
They provide the data to all different offices, right?
you
var
That's a real competition.
As I said, one is official statistics.
med.
And one is kind of big data, right?
But obviously, as the professor mentions,
Official statistics are credible, you know.
trust, transparency, all those things.
But we need to look forward.
I think those are classic definitions you have.
You own it, but you need to go further.
I would say there are two steps.
The first step for the official statistics,
Obviously you have credentials, you have trust,
You keep doing a great job.
People depend on your information, that's for sure.
But the first step is,
I think it's important for official statistics bureaus
or whatever in whatever countries to adapt artificial intelligence
to use big data to improve the efficiency of your process of making
in a faster, more productive way, with broader coverage,
granularities, bring more information, details, and construction.
A truly solid foundation for your credentials, trust, and transparency.
Yes, thank you.
But second, no, let me go to the second one.
I think the second one is more important.
I'm sorry, I'm sorry, this is the boss.
I was in your hands.
No, the second one is more important.
Yeah.
Because there is competition, right?
I mean, I say big data is competing with you, trying to eat your cake.
So you need to be brave, dive into big data,
and try to learn how to do those things.
This is very important for everyone in this room.
Assume 100 years later, so today is 2126.
Assume today is 2626, 100 years later,
You're always living in that time.
We have unlimited data for free.
We have unlimited computing power for free.
Will big data be able
to mimic all your national accounts?
in a far more effective way, with a significantly higher accuracy rate?
We don't know, but it's possible, isn't it?
You can't do it today, no one can do it today,
because no one has unlimited data,
No one has unlimited computing power.
So if that day comes, you'll have to think about it
who has the authority to define the results,
Who has the public trust to classify the variable?
And bring transparency to the entire world.
So I think that's the official statistics for today.
Try to learn from this big data for the future.
Because I don't know whether the future is official or big data.
I think two things may emerge.
May, if unlimited data, unlimited computing power,
Maybe it's big data that will overtake everything.
But still, we need someone, which I think will be official statistics.
to provide the framework, to provide the concept
to provide the classifications
So I think this second part is really important.
So you will need to understand statistics in the future,
A hundred years later, and come back to start today,
To build the path, move into the future.
In that sense, I think...
This is important. I'm sorry for taking up your time.
No, no, you said very wise things,
And I see a lot of nodding heads in the audience.
So, we statisticians provide standards, methodologies, and concepts,
and if you follow those, the data are well understood,
they are comparable across countries over time,
And that is truly necessary.
And we should use big data as inputs for that.
And apart from that, you can use fast big data.
Do you have a budget for computing power today?
Well, everyone in the world struggles.
with the rising cost of tokens,
But we use big data, we use AI—yes, at the IMF we do.
That's really an important role—we set the standards.
And actually today we had a presentation.
by someone from the Central Bank of Nigeria
Who said, well, this AI adoption,
We obtained all this information from OpenAI.
from Anthropic, from Microsoft, but none of it is comparable,
So how does it add up? How do we arrive at a complete picture?
And our role as statisticians is to establish standards, taxonomies,
and see whether we can form a shared understanding of these matters,
and using big data.
Wow, that's fascinating.
In addition to what Brian said, you can use fast indicators.
to have a quick indication of what's going on.
My concern is more about multi-distribution sites,
because we always have trouble defining the standards,
Collect the data, generate the data, display it on the website,
but then people go to the LLMs, to the clouds and the ChatGPTs,
They type in questions and receive data.
from somewhere, not specifically from ours...
So that's more of a challenge I see,
But we will work together with technology companies to see what we can do.
We have to go to the last question.
Which is the one million dollar question I'm going to ask you about.
What should we, official statistics stations,
What should we take from this discussion as our first practical priority?
Brian
I think the first practical priority,
In my view, it should be to think more about
How to incorporate these big data and AI tools in a way that does not
undermine the public's trust in the credibility of the institutions.
So, as I said, I think it's very useful that there's going
to be private sector data products that offer a certain symbiosis that
have reinforcing effects through the competition they generate.
But official statistical agencies definitely still have a role,
and that role is to provide a certain level of credibility,
a certain level of institutional continuity
And so finding a way of threading that needle between adopting
these new tools in a way that helps resolve some
of the issues that we have highlighted here today.
But also not forgetting that there is a role
for moving with some degree of foresight,
moving with some level of caution about,
Well, we need to test these things.
in such a way that we believe the public
will maintain confidence in the credibility of official statistics.
So threading that needle, I think, is a crucial thing.
that official statistical agencies should be considering.
Thank you. Very good.
Yeah, I
Follow the same line.
I think the most urgent issue is for the official statistical agencies.
think
to jump into the water, securing the funds for computing power,
Jørgen
Try to use big data to perform simulations.
to try to perform specific tasks, and to try to compare.
Whatever big data generates for you, and whatever your classical,
You know, the national accounts provide you with
and see what similarities there are, what differences there are,
I mean the actual transparency and the classification
and the quantification of those issues, then you begin to build
a framework to think about how you can move into the future
Then you begin to bring everyone in the world together.
And then we will have a new world.
which is to combine classical statistics
and big data and AI there will be a
wonderful world in 2126. thank you, Min. I think you're glad you're
So aligned on this
I think we've reached the end of the panel.
Thank you so much.
It was really great to hear your perspectives.
from your respective backgrounds
on what we statisticians should do
I think the challenge you gave us
is aligned with what we want to accomplish.
It makes our lives very exciting.
I mean, we love using all these new data sources.
and generate statistics from it.
And statistics are based on methods,
concepts, standards, methodologies, etc.
So they can be trusted, and we must explain that as well.
So, thank you for that.
I would like to ask for a round of applause for our panelists.
No, you can sit down.
You may sit down.
I'm going to close the panel.
Not only the panel, I'm going to shut down the entire conference.
It's also always a sorrowful moment, as I've already mentioned.
Because I truly loved those two days.
I think you, thank you Brian and Min.
It was a very fitting conversation.
which to close this 14th IMF Statistical Forum.
It brought us back to the question I posed at the beginning.
Where do we want our statistical system to be five years from now?
So over the past two days,
We have examined economic measurements from various perspectives.
We have seen how big data, new data sources, and AI, SI,
can make information more timely, more granular, and reveal aspects
of the economy that were previously difficult to observe.
And we have asked what this means in practice,
both for economic policymakers,
for quality and for the institutions that produce official statistics.
And if there is one message I take away from these discussions,
it is that our objective is not more data for the sake of more data,
It provides better evidence for better decisions.
And getting there, I believe it comes down to three things:
and one purpose that binds them together.
First, economic measurement is required
to keep pace with the changing economy.
Across the forum, we have seen how new data sources
and AI are expanding what we can observe,
How quickly we can observe it, and at what level of detail.
And the challenge now is to turn
Transform these possibilities into statistics and insights
that are genuinely useful to policymakers
while being equally clear about the limitations.
And as I asked at the outset,
It means being willing to ask what we should do differently.
And we received clear guidance from our panelists on that today.
And the second, trust remains fundamental.
As the data ecosystem becomes increasingly complex,
quality, transparency, comparability
Confidentiality and methodological rigor become even more important.
The tools and sources may change,
However, the credibility of official statistics remains essential.
That is our unique role in the information landscape,
producing statistics based on well-defined concepts and methodologies
that are comparable over time and across countries.
We talked a lot about communication.
Communication with our users is key,
about our methods and metadata,
the strengths and limitations of our data,
the assumptions we make
and how our statistics can and cannot be used.
And we should be proud of the contribution.
Official statistics contribute to the information landscape.
And we should work to ensure that they can
be produced independently in each country,
with sufficient resources and access to the information required
to do the job and the computational power.
The IMF will continue
to champion strong statistical institutions in all countries.
A third collaboration is essential.
No single institution has all the data,
expertise or capabilities required.
Statistical organizations, central banks,
international organizations, researchers, politicians, and
The private sector brings a different perspective.
Today we have seen how much we can learn from each other.
how much these collaborations mean
more and more in our statistical family.
The purpose of it all is better evidence for better decisions.
The more timely and detailed information there is, the more helpful it is.
To better understand the economy, and ultimately support better policy.
That's why we must keep strengthening the connection between
those who produce statistics and those who use them.
We need to talk to policymakers about what information they need.
What hinders them from obtaining the correct information?
how they can make better decisions.
So perhaps the thought I would leave you with is this.
The future of economic measurement
It is not about replacing what we have built.
It's about building on the strength of official statistics,
while making the best use of new data,
new methods and new capabilities
And if we can achieve that while maintaining trust
and keeping the needs of policymakers and users at the center,
we will be much better equipped to measure
an economy that is changing even more quickly.
And that, I hope, gives us a good start on the question we began with.
Where do we want to be five years from now?
So, before we close, let me thank all of our speakers,
moderators, presenters, and panelists
for sharing their expertise and perspectives so openly.
As I said in my opening remarks,
We need everyone's input to ensure we find the right way forward
as a statistical community and provide the greatest value.
In the coming weeks, we will look back on everything that has been said.
And we will incorporate it into our future plans.
Now, thank you all, to you all who joined us here in Washington.
and online for your questions, ideas, and engagement.
It's wonderful to see so many of you engaging actively and providing us with feedback.
Last but not least, I would like to thank my colleagues across the IMF.
who frequently worked
And worked tirelessly behind the scenes
to make these two days possible
I know how much effort went into this.
in preparing it, delivering this forum.
I'm truly so proud of all of you—you did an outstanding job.
So Alessandra, Andrew, Annika, Bindu, Binta, Christina,
Claudia, Dane, Deepti, Francine, Jim, Margarida
Michael, Neil, Niva, Oyin, Xi, Rukaiya, Priyata, and Yolanda
Please stand up and let's give them a round of applause.
Please stand up.
Thank you.
You're amazing.
Okay, I'm sure we all leave this forum with new ideas, new connections,
and a shared commitment to continue this conversation.
Thank you very much, and I wish everyone a safe journey home.
Thank you.
