Popular in Economics

Economics

Gender and Economics

University of California Santa Barbara Shelly Lundberg

Banks and Policy Organisations

Greening the financial sector

European Investment Bank Matteo Ferrazzi

Economic History

Gravity with Gravitas: A Solution to the Border Puzzle

Boston College James E. Anderson

Economic Inequalities

Visions of Inequality

City University of New York Branko Milanovic

Economics

School Finance Equalization Increases Intergenerational Mobility

Yale University Barbara Biasi

Economics

Olivier Blanchard on U.S. Pandemic-Era Inflation

Peterson Institute for International Economics Olivier Blanchard

Econometric Modelling of the Aggregate Time Series Relationship Between Consumers Expenditure and Income in the UK

View more insights in Most Cited - Social Sciences - Economics - Econometrics

View more figures from University of Exeter

View the page of James Davidson

Read the Study

Simple time‐series representations dominated quarterly permanent‐income/life‐cycle models of consumption in fit and predictive accuracy. However, an ‘error‐correction’ model (ECM, using the log consumption/income ratio) reconciled both the theories and the evidence, and treated as the DGP, explained the connection between the time‐series and econometric equations.

Image courtesy of interviewee. December 4, 2023

Generating Workspace

Export as PDF

The speaker reflects on a paper they wrote 40 years ago while still a student, which is associated with their master's thesis. The paper gained notoriety for addressing non-stationary time series modeling, revealing a correct approach that gained attention in the research community—building upon prior, less recognized work. The paper primarily focused on modeling economic relationships between consumption, income, and other variables.

A key contribution was the concept of cointegration, originally popularized by Clive Granger, which allows non-stationary variables to be analyzed by transforming them into stationary combinations. The speaker's study compared various methodologies and highlighted discrepancies in results among different research approaches, particularly emphasizing the importance of using both levels and differences of non-stationary variables in modeling.

The paper suggested using growth rates and included economic theory, particularly the savings ratio, in its regression analysis. While initially seen as ad hoc, the approach ultimately showed robust predictive power over the years, maintaining relevance even as new data emerged.

The speaker notes that while their original understanding was limited, subsequent developments in macroeconomic research significantly expanded the framework of cointegration, leading to its acceptance as a crucial tool for analyzing economic data.

In summary, the discussion emphasizes the evolution of thinking about time series data in economics, the enduring impact of the speaker's research, and the broader implications for macroeconomic modeling techniques in light of the rational expectations revolution.

Language:

Export as PDF

Export as PDF
It appears that you may have intended to provide a specific video for me to analyze, but it wasn't included in your message. Please provide the video you would like me to summarize or analyze, and I'll be happy to help!

Export as PDF
It seems like you've two periods ('..') without specifying the video or convideo you're referring to. Could you please provide more details or clarify the video you're interested in? Once I have that information, I can help identify key quotes or themes from it.

Export as PDF
Here are some similar articles or topics that resonate with the themes and concepts presented in the video:

1. **Cointegration and Economic Time Series Analysis**:
- Articles discussing the development and application of cointegration in economic modeling, particularly those referencing the works of Clive Granger and his contributions to econometrics.

2. **Non-Stationary Time Series Models**:
- Research papers focused on the methodologies for modeling non-stationary time series data, including various techniques to address issues related to spurious regression, and the importance of differentiating between levels and differences.

3. **Historical Perspective on Macroeconomic Modeling**:
- Discussions on the historical evolution of macroeconomic models, including foundational ideas pre- and post-rational expectations revolution, explaining how these shifts have impacted current modeling practices.

4. **Comparative Methodologies in Time Series Analysis**:
- Studies that compare the effectiveness of different methodologies (OLS, ARIMA, VAR, etc.) in modeling economic variables, similar to how the speaker’s paper highlighted varying results from different approaches.

5. **The Savings Ratio and Economic Growth**:
- Research articles that delve into the relationship between savings ratios, consumption, and income, with a focus on their predictive capabilities in economic forecasts and growth theories.

6. **Advances in Macroeconomics: Rational Expectations and Beyond**:
- Papers discussing the implications of the rational expectations hypothesis on historical and contemporary economic models, elaborating on how economic agents’ expectations adjust to new information.

7. **Art of Predictive Modeling in Economics**:
- Articles that explore the art and science behind building predictive models in economics, emphasizing the importance of robust methodologies, model validation, and the role of theory in guiding empirical analysis.

8. **Evolution of Cointegration Techniques**:
- A review of how cointegration techniques have evolved since their inception, exploring new developments, applications, and extensions in various economic convideos.

By exploring these articles or topics, one can gain a deeper understanding of the themes related to time series modeling, cointegration, and their implications in macroeconomic research.

X-TIMESTAMP-MAP:LOCAL=00:00:00.000,MPEGTS=900000

00:00.135 --> 00:03.105
This paper was written 40 years ago, Joll early

00:03.655 --> 00:04.865
when I was still a student.

00:05.285 --> 00:08.945
At least in part not the work was done as part

00:08.945 --> 00:11.545
of my master's thesis, actually, or part of the work,

00:11.825 --> 00:14.505
although, it didn't come out in 1978.

00:14.625 --> 00:15.945
I graduated in 1975.

00:16.925 --> 00:19.025
So I've had a bit of fun rereading this

00:19.035 --> 00:22.065
after many, many years at your suggestion

00:22.085 --> 00:23.505
and thinking, what did we do there?

00:23.805 --> 00:27.145
And, of course, it is a bit notorious as this,

00:27.535 --> 00:30.945
this paper, and I can tell you why it's notorious,

00:31.065 --> 00:32.305
I think in, in a few words.

00:33.165 --> 00:37.595
Um, it's the, it, it really did for the first time,

00:38.095 --> 00:39.315
not quite the first time,

00:39.315 --> 00:43.235
because my other teacher, Dennis Saga, because my co-author

00:43.255 --> 00:46.275
and mentor was David Hendry with this paper at,

00:46.275 --> 00:47.955
this is at the London School of Economics.

00:48.055 --> 00:51.155
Dennis Sagan had something similar in the paper 10 years

00:51.475 --> 00:53.235
previously, but nobody really noticed it.

00:53.235 --> 00:57.035
It was asked that caught the attention of the,

00:57.375 --> 00:58.515
the research community.

00:59.365 --> 01:02.825
And this was to discover the correct way

01:02.925 --> 01:06.345
of modeling non stationary time series,

01:07.455 --> 01:10.625
because it had been done in the wrong way for,

01:10.885 --> 01:11.945
you know, quite a while.

01:12.125 --> 01:15.105
And one of the things that our paper does is

01:15.105 --> 01:18.985
to look at various examples of consumption, income

01:19.495 --> 01:21.145
regressions,

01:21.295 --> 01:24.425
with various dynamic aspects built into them.

01:25.005 --> 01:28.065
Of course, the, the person whose name one should mention,

01:28.065 --> 01:30.585
because he's, he is the one that got the Nobel Prize.

01:30.585 --> 01:34.625
He's Clive Granger because he, invented

01:35.165 --> 01:37.705
and popularized the notion of cointegration.

01:38.445 --> 01:40.145
And basically that's what we were doing.

01:40.285 --> 01:42.545
We were doing cointegration, but we didn't know it.

01:43.085 --> 01:47.185
All we knew was that we found an equation which,

01:47.185 --> 01:50.585
contained rates of change of the, in other words,

01:50.685 --> 01:54.425
the differences from year to year of our variables,

01:54.455 --> 01:58.305
consumption and income and inflation, and the price level

01:58.405 --> 02:01.745
and things of that kind as well as the levels.

02:01.925 --> 02:03.425
Of course, the levels are trending.

02:03.485 --> 02:05.985
The levels are aren't staying in one place.

02:05.985 --> 02:07.665
They're not, they're non stationary.

02:08.205 --> 02:11.665
But what five grains of notice was

02:11.785 --> 02:14.545
that if you look at linear combinations

02:14.685 --> 02:17.465
of non-stationary variables,

02:18.125 --> 02:20.385
the result can be a stationary process and

02:20.385 --> 02:23.585
therefore can be put into the same equation.

02:24.285 --> 02:27.225
Now, an something which I'd rather forgotten about,

02:27.225 --> 02:30.665
but it's a very important aspect of the paper, is

02:30.665 --> 02:32.945
that it compares different studies which

02:33.565 --> 02:36.545
are looking at the same, same times zero.

02:36.545 --> 02:39.665
It is the same problem with the same sort of economic theory

02:40.245 --> 02:41.945
in their minds,

02:41.965 --> 02:44.425
and yet really getting rather different results.

02:45.185 --> 02:49.405
And the particular paper, which was a, a curiosity to us,

02:50.545 --> 02:53.885
was, I forget the title

02:53.945 --> 02:56.525
and the names of the authors wall was one of them

02:56.525 --> 02:59.005
because that's how he is identified in, in, in,

02:59.265 --> 03:01.045
in our paper as wall.

03:01.045 --> 03:02.125
At wall at al.

03:02.615 --> 03:05.965
These people had been reading a, an important book

03:06.145 --> 03:09.605
by two chaps called Boxing Jenkins, who were not economists.

03:09.605 --> 03:13.505
They were concerned with modeling,

03:13.695 --> 03:17.265
statistical modeling of time series for purposes such

03:17.265 --> 03:18.585
as forecasting and,

03:19.205 --> 03:23.945
and, what they were ba their, their, their philosophy,

03:24.155 --> 03:27.385
which was really important as merging about the same sort

03:27.385 --> 03:32.035
of time, was that you could only really construct

03:32.685 --> 03:36.055
relationships between variables when they're all stationary.

03:36.985 --> 03:40.085
In other words, when they, when they have a a,

03:40.245 --> 03:43.765
a fixed distribution, they don't wander off in the same way

03:43.765 --> 03:45.525
that, things like stock prices

03:45.905 --> 03:49.125
and, GDP, you know, these things grow over time.

03:49.515 --> 03:52.485
Well, that doesn't do, you can't build,

03:53.045 --> 03:55.085
although that's what of course, economists were attempting

03:55.085 --> 03:58.285
to do and still do, attempt to do, you know, at times.

03:58.865 --> 04:01.765
But that we need to model stationary time series.

04:02.145 --> 04:04.085
And if they're not stationary, when we,

04:04.155 --> 04:05.485
when we receive the data,

04:05.875 --> 04:07.525
well then we don't look at the level,

04:07.625 --> 04:09.885
we look at the change from period to period,

04:09.885 --> 04:12.365
because that will usually appear stationary.

04:13.105 --> 04:15.805
And so, you know, the, the, these,

04:16.055 --> 04:19.285
these researchers were following, this,

04:19.795 --> 04:22.605
this line in which they were looking at the differences,

04:22.865 --> 04:24.925
the growth rates of consumption

04:25.065 --> 04:27.445
and the growth rate of income,

04:27.665 --> 04:29.845
and showing indeed that they were closely related.

04:30.665 --> 04:34.125
But of course, their relationship can't really tell you very

04:34.125 --> 04:37.125
much about what's going on in the heads of consumers

04:37.125 --> 04:38.965
because the levels have been taken out.

04:38.965 --> 04:41.845
They have difference. So they're following a statistical,

04:42.265 --> 04:44.485
optimality, path,

04:45.065 --> 04:48.215
and yet they're throwing away any contribution

04:48.215 --> 04:49.215
of economic theory.

04:49.995 --> 04:54.135
The puzzle was that their, their model seemed

04:54.135 --> 04:56.175
to fit the data better than all the models,

04:56.185 --> 04:58.535
which incorporated economic theory by,

04:58.835 --> 05:01.055
by having good savings ratios

05:01.115 --> 05:04.015
and, relationship between consumption

05:04.075 --> 05:08.135
and income, as, part of them.

05:09.185 --> 05:11.405
So what, where did that leave us?

05:11.735 --> 05:16.055
Where it left us was that we thought we've gotta

05:17.305 --> 05:19.915
basically stick to this important notion

05:19.945 --> 05:23.355
that we can only really model stationary variables using,

05:23.445 --> 05:24.835
using regression techniques.

05:25.105 --> 05:26.395
Okay? Um,

05:26.975 --> 05:31.225
but on the same time, we want to get the levels

05:32.485 --> 05:35.145
of the variables as well as the differences

05:35.165 --> 05:36.985
of the variables into the equation.

05:37.205 --> 05:40.705
How did we do that? And in fact, the truth is that this was,

05:41.125 --> 05:44.465
I recall when I did this in my master's thesis,

05:44.545 --> 05:45.745
I was very apologetic.

05:45.985 --> 05:49.425
I said, this is totally ad hoc and stupid,

05:49.605 --> 05:51.505
but look, if I put the difference,

05:51.645 --> 05:52.945
and this is in logarithms,

05:52.965 --> 05:54.825
so it's really the log of a ratio.

05:55.525 --> 05:58.165
If I, if I put the difference of the logs

05:58.165 --> 06:01.525
of these two variables, well, with a, you know, last year,

06:01.585 --> 06:05.405
you know, we put a lag on so that, so that we have the,

06:05.425 --> 06:08.405
you know, the, the direction of causation appropriate.

06:09.195 --> 06:11.815
Uh, then the difference looks stationary

06:12.635 --> 06:14.415
and the changes are stationary.

06:14.415 --> 06:16.815
So the whole equation fits the box

06:16.815 --> 06:21.095
and Jenkins, prescription

06:21.235 --> 06:23.255
of best fitting stationary data.

06:24.085 --> 06:25.565
Um, and

06:25.665 --> 06:29.645
yet we've included the savings ratio in the equation, which

06:29.645 --> 06:31.845
therefore we expect to have some predictive power.

06:32.465 --> 06:36.245
Um, if, if it, it's, if, if, you know, the way

06:36.355 --> 06:41.285
that consumers behave informs how these data came to be.

06:42.395 --> 06:45.495
Um, and so I think that's what the, you know, what is known

06:45.495 --> 06:47.975
as the daisy paper, 'cause it's DHS

06:47.975 --> 06:51.385
and Y, the authors, of whom I'm the D

06:51.445 --> 06:53.265
and H is David Henry.

06:53.885 --> 06:57.585
Um, the other guys were only involved in the revision

06:57.725 --> 07:00.385
to do some extra data, you know, some extra,

07:00.455 --> 07:01.465
data analysis.

07:01.565 --> 07:04.265
So, I think we have to say the, the first version

07:04.285 --> 07:06.225
of this paper was just Davidson and Henry.

07:06.405 --> 07:09.305
So that's, that's how it all came, came about.

07:10.445 --> 07:14.225
Um, but what we were basically doing was saying, okay,

07:14.285 --> 07:17.545
you know, we, we've, we've, we've resolved,

07:18.315 --> 07:19.945
we've explained the difference

07:19.945 --> 07:22.185
between these different studies by saying,

07:22.765 --> 07:25.905
got on the one hand, follow the, you know, the,

07:25.905 --> 07:27.345
the box of Jenkins prescription.

07:27.445 --> 07:29.945
On the other hand, get the theory in there

07:30.125 --> 07:31.385
by getting the levels and

07:31.385 --> 07:32.465
the relationship between the levels.

07:33.245 --> 07:36.535
But reading that paper, in retrospect, of course,

07:36.555 --> 07:39.175
we had no idea really what we were doing,

07:39.175 --> 07:42.015
because Clive Granger hadn't yet read it.

07:42.875 --> 07:44.815
And the con, you know, the relevant concepts

07:44.815 --> 07:46.055
didn't, didn't yet exist.

07:46.195 --> 07:50.135
And this was, what amazingly grew over

07:50.955 --> 07:52.135
the next 20 years

07:52.235 --> 07:56.505
or so of macroeconomics research,

07:57.035 --> 07:59.745
which was the whole, the whole notion of cointegration

07:59.775 --> 08:01.745
that you, you, you look at collections

08:01.745 --> 08:03.265
of non-stationary variables.

08:03.265 --> 08:06.185
You don't ask if they're correlated

08:06.815 --> 08:09.185
because that's a notion which is only really

08:09.585 --> 08:10.905
relevant to stationary data.

08:11.445 --> 08:13.465
You ask if they're cointegration, which means

08:13.465 --> 08:14.625
that they have common trends.

08:15.205 --> 08:18.705
If you take suitable linear combinations of them,

08:19.005 --> 08:22.665
you get a stationary series, which can then be correlated,

08:22.875 --> 08:27.225
which is not co correlated with the growth with the,

08:27.385 --> 08:29.585
with the variables measuring differences

08:29.585 --> 08:30.625
from period to period.

08:31.545 --> 08:34.705
Um, now our equation

08:35.965 --> 08:38.985
did roll well because it turned out that,

08:39.365 --> 08:41.905
it fitted the data pretty well, but it was quite stable.

08:42.145 --> 08:45.265
I, I, I, there were subsequent studies oath

08:45.325 --> 08:48.905
for at least the next 15 years, which really found,

08:49.365 --> 08:51.825
you know, in other words, with new data that hadn't been,

08:52.855 --> 08:55.325
hadn't occurred when we wrote the original play.

08:55.325 --> 08:57.005
But going up into the 1980s,

08:57.785 --> 08:59.125
and it still fitted pretty well.

08:59.125 --> 09:00.245
It broke down eventually.

09:00.405 --> 09:03.045
I forget exactly when, sometime in the nineties

09:03.075 --> 09:04.285
that it collapsed.

09:04.745 --> 09:09.245
Um, but, so what this was telling us

09:09.245 --> 09:13.525
that we were isolating some,

09:14.905 --> 09:18.665
regularities in the data, that it wasn't the kind

09:18.665 --> 09:20.705
of regression which breaks down as soon

09:20.705 --> 09:22.305
as you add a new observation

09:22.895 --> 09:24.505
that hadn't been used to fit it.

09:24.645 --> 09:26.545
And of course, this is the, this is the problem

09:26.575 --> 09:28.585
with many regressions is that they, you know,

09:28.585 --> 09:30.105
they fit the sample well enough

09:30.105 --> 09:33.665
because you fix them to do so, put in one variable

09:33.665 --> 09:36.025
or another, but as soon as you bring in some new data,

09:36.135 --> 09:38.065
then you know, the whole thing falls apart.

09:38.495 --> 09:41.865
Well, the daisy consumption function

09:42.645 --> 09:44.425
really did rather well in that respect.

09:44.965 --> 09:49.025
Um, and, and it was u it was widely used, for,

09:49.045 --> 09:51.705
you know, by forecasters something, something similar.

09:51.925 --> 09:54.065
The, the whole ever correction idea.

09:54.785 --> 09:57.025
'cause the, where the ever correction,

09:57.095 --> 09:59.625
terminology comes from is precisely

09:59.625 --> 10:04.105
that you've got the savings ratio as it lags, as one

10:04.105 --> 10:05.385
of the explanatory variables.

10:05.385 --> 10:08.025
So if this moves, if people find that the

10:08.585 --> 10:10.385
saving more than they,

10:10.735 --> 10:13.065
than they do normally do in the long term,

10:13.735 --> 10:16.505
because their incomes have hopefully risen

10:16.505 --> 10:19.025
and their consumption has remained according

10:19.025 --> 10:20.625
to their usual habits and,

10:20.725 --> 10:24.625
and, so, golly, look, we've,

10:24.625 --> 10:26.865
we've got all this cash, let's go out

10:26.865 --> 10:27.905
and spend it on something.

10:27.965 --> 10:29.905
But this is something that doesn't happen immediately.

10:29.965 --> 10:32.625
It happens with an, with some irregular lag.

10:33.045 --> 10:36.345
So, so hence the error correction is that we assume

10:36.345 --> 10:39.065
that people focus on, of course, this doesn't only apply

10:39.065 --> 10:41.705
to consumption, it applies to lots of things, you know,

10:41.705 --> 10:45.225
money stock and, and, and, and,

10:45.225 --> 10:49.185
likewise in financial data, of course, there's also

10:49.925 --> 10:53.625
an awful lot of research has been done over the 40 years

10:53.775 --> 10:57.825
following, our original contribution

10:58.365 --> 11:00.625
in which, you know, this, this phenomenon

11:00.625 --> 11:03.065
of cointegration has turned up in a lot of places.

11:03.965 --> 11:06.745
And so it's been regarded as, as,

11:06.805 --> 11:09.185
as quite an important contribution to

11:10.015 --> 11:13.305
macroeconomic research, a great deal of the effort

11:13.375 --> 11:16.185
that took place in the latter half of the 20th century,

11:16.285 --> 11:20.145
you could say, into, into macroeconomic modeling,

11:20.145 --> 11:25.025
rather collapsed because it was pointed out that,

11:25.365 --> 11:29.865
that expectations, had to be modeled in terms

11:29.885 --> 11:32.585
of a complete,

11:32.975 --> 11:36.345
macroeconomic system rather than, you know,

11:36.345 --> 11:38.225
being simple, simple projections and,

11:38.445 --> 11:41.185
and essentially that, that, that, that large

11:41.815 --> 11:44.625
macroeconomic models, weren't expected

11:44.645 --> 11:47.225
to work very well, which we didn't anticipate, I think,

11:47.225 --> 11:48.705
at the time that we wrote this paper.

11:48.925 --> 11:52.905
So, so, so what, what is known as the rational expectations

11:53.325 --> 11:56.445
Revolution, rather shot a hole in some of this work.

11:56.985 --> 11:58.365
But, I think

11:58.365 --> 12:03.125
nonetheless, cointegration has continued to be a,

12:03.905 --> 12:06.645
you know, an, you know, an important tool in the Armory

12:07.025 --> 12:10.365
for understanding, non-stationary variables.

00:00 - This paper was written 40 years ago, Joll early when I was still a student. At least in part not the work was done as part of my master's thesis, actually, or part of the work, although, it didn't come out in 1978. I graduated in 1975.

00:16 - So I've had a bit of fun rereading this after many, many years at your suggestion and thinking, what did we do there? And, of course, it is a bit notorious as this, this paper, and I can tell you why it's notorious, I think in, in a few words. it's the, it, it really did for the first time, not quite the first time, because my other teacher, Dennis Saga, because my co-author and mentor was David Hendry with this paper at,

00:46 - this is at the London School of Economics. Dennis Sagan had something similar in the paper 10 years previously, but nobody really noticed it. It was asked that caught the attention of the, the research community. And this was to discover the correct way of modeling non stationary time series, because it had been done in the wrong way for, you know, quite a while. And one of the things that our paper does is

01:15 - to look at various examples of consumption, income regressions, with various dynamic aspects built into them. Of course, the, the person whose name one should mention, because he's, he is the one that got the Nobel Prize. He's Clive Granger because he, invented and popularized the notion of cointegration. And basically that's what we were doing. We were doing cointegration, but we didn't know it. All we knew was that we found an equation which,

01:47 - contained rates of change of the, in other words, the differences from year to year of our variables, consumption and income and inflation, and the price level and things of that kind as well as the levels. Of course, the levels are trending. The levels are aren't staying in one place. They're not, they're non stationary. But what five grains of notice was that if you look at linear combinations of non-stationary variables,

02:18 - the result can be a stationary process and therefore can be put into the same equation. Now, an something which I'd rather forgotten about, but it's a very important aspect of the paper, is that it compares different studies which are looking at the same, same times zero. It is the same problem with the same sort of economic theory in their minds, and yet really getting rather different results. And the particular paper, which was a, a curiosity to us,

02:50 - was, I forget the title and the names of the authors wall was one of them because that's how he is identified in, in, in, in our paper as wall. At wall at al. These people had been reading a, an important book by two chaps called Boxing Jenkins, who were not economists. They were concerned with modeling, statistical modeling of time series for purposes such as forecasting and,

03:19 - and, what they were ba their, their, their philosophy, which was really important as merging about the same sort of time, was that you could only really construct relationships between variables when they're all stationary. In other words, when they, when they have a a, a fixed distribution, they don't wander off in the same way that, things like stock prices and, GDP, you know, these things grow over time. Well, that doesn't do, you can't build, although that's what of course, economists were attempting

03:55 - to do and still do, attempt to do, you know, at times. But that we need to model stationary time series. And if they're not stationary, when we, when we receive the data, well then we don't look at the level, we look at the change from period to period, because that will usually appear stationary. And so, you know, the, the, these, these researchers were following, this, this line in which they were looking at the differences,

04:22 - the growth rates of consumption and the growth rate of income, and showing indeed that they were closely related. But of course, their relationship can't really tell you very much about what's going on in the heads of consumers because the levels have been taken out. They have difference. So they're following a statistical, optimality, path, and yet they're throwing away any contribution of economic theory.

04:49 - The puzzle was that their, their model seemed to fit the data better than all the models, which incorporated economic theory by, by having good savings ratios and, relationship between consumption and income, as, part of them. So what, where did that leave us? Where it left us was that we thought we've gotta basically stick to this important notion that we can only really model stationary variables using,

05:23 - using regression techniques. Okay? but on the same time, we want to get the levels of the variables as well as the differences of the variables into the equation. How did we do that? And in fact, the truth is that this was, I recall when I did this in my master's thesis, I was very apologetic. I said, this is totally ad hoc and stupid, but look, if I put the difference,

05:51 - and this is in logarithms, so it's really the log of a ratio. If I, if I put the difference of the logs of these two variables, well, with a, you know, last year, you know, we put a lag on so that, so that we have the, you know, the, the direction of causation appropriate. then the difference looks stationary and the changes are stationary. So the whole equation fits the box and Jenkins, prescription

06:21 - of best fitting stationary data. and yet we've included the savings ratio in the equation, which therefore we expect to have some predictive power. if, if it, it's, if, if, you know, the way that consumers behave informs how these data came to be. and so I think that's what the, you know, what is known as the daisy paper, 'cause it's DHS and Y, the authors, of whom I'm the D and H is David Henry.

06:53 - the other guys were only involved in the revision to do some extra data, you know, some extra, data analysis. So, I think we have to say the, the first version of this paper was just Davidson and Henry. So that's, that's how it all came, came about. but what we were basically doing was saying, okay, you know, we, we've, we've, we've resolved, we've explained the difference between these different studies by saying,

07:22 - got on the one hand, follow the, you know, the, the box of Jenkins prescription. On the other hand, get the theory in there by getting the levels and the relationship between the levels. But reading that paper, in retrospect, of course, we had no idea really what we were doing, because Clive Granger hadn't yet read it. And the con, you know, the relevant concepts didn't, didn't yet exist.

07:46 - And this was, what amazingly grew over the next 20 years or so of macroeconomics research, which was the whole, the whole notion of cointegration that you, you, you look at collections of non-stationary variables. You don't ask if they're correlated because that's a notion which is only really relevant to stationary data. You ask if they're cointegration, which means

08:13 - that they have common trends. If you take suitable linear combinations of them, you get a stationary series, which can then be correlated, which is not co correlated with the growth with the, with the variables measuring differences from period to period. now our equation did roll well because it turned out that, it fitted the data pretty well, but it was quite stable. I, I, I, there were subsequent studies oath

08:45 - for at least the next 15 years, which really found, you know, in other words, with new data that hadn't been, hadn't occurred when we wrote the original play. But going up into the 1980s, and it still fitted pretty well. It broke down eventually. I forget exactly when, sometime in the nineties that it collapsed. but, so what this was telling us that we were isolating some,

09:14 - regularities in the data, that it wasn't the kind of regression which breaks down as soon as you add a new observation that hadn't been used to fit it. And of course, this is the, this is the problem with many regressions is that they, you know, they fit the sample well enough because you fix them to do so, put in one variable or another, but as soon as you bring in some new data, then you know, the whole thing falls apart.

09:38 - Well, the daisy consumption function really did rather well in that respect. and, and it was u it was widely used, for, you know, by forecasters something, something similar. The, the whole ever correction idea. 'cause the, where the ever correction, terminology comes from is precisely that you've got the savings ratio as it lags, as one of the explanatory variables. So if this moves, if people find that the

10:08 - saving more than they, than they do normally do in the long term, because their incomes have hopefully risen and their consumption has remained according to their usual habits and, and, so, golly, look, we've, we've got all this cash, let's go out and spend it on something. But this is something that doesn't happen immediately. It happens with an, with some irregular lag.

10:33 - So, so hence the error correction is that we assume that people focus on, of course, this doesn't only apply to consumption, it applies to lots of things, you know, money stock and, and, and, and, likewise in financial data, of course, there's also an awful lot of research has been done over the 40 years following, our original contribution in which, you know, this, this phenomenon of cointegration has turned up in a lot of places. And so it's been regarded as, as,

11:06 - as quite an important contribution to macroeconomic research, a great deal of the effort that took place in the latter half of the 20th century, you could say, into, into macroeconomic modeling, rather collapsed because it was pointed out that, that expectations, had to be modeled in terms of a complete, macroeconomic system rather than, you know, being simple, simple projections and, and essentially that, that, that, that large

11:41 - macroeconomic models, weren't expected to work very well, which we didn't anticipate, I think, at the time that we wrote this paper. So, so, so what, what is known as the rational expectations Revolution, rather shot a hole in some of this work. But, I think nonetheless, cointegration has continued to be a, you know, an, you know, an important tool in the Armory for understanding, non-stationary variables.

Faculti AI can make mistakes