In Part 1, we looked at the ratio of consumption spending to net worth, and how it changed over time. This time we’ll look at the correlation between net worth and consumption.
Here is the big picture: personal consumption expenditures (FRED Series PCE) plotted against Net Worth (FRED series TNWBSHNO) Data is per calendar quarter.
Graph 1 Consumption vs Net Worth, 1959 – 2011
The data set is divided into two segments, with a break point at the beginning of 1997, with net worth at $30,315 and consumption at $5,467. In a close up view, there is a clear slope change there. Still, the selection is a bit arbitrary, since the high point of Q3, 1994, could also have been chosen, with net worth at $25291, and PCE at $4856. But that is a detail, and no other reasonable breaks stand out.
Notably, both the slope of the data line and R^2 are significantly less after the break. Visually, it’s obvious that in the later data, there is a lot more scatter. Also note that that big data moves post 1997 return to the continuation of the best fit line, pre-1997. Slope and R^2 measurements for the entire data set and the two segments are presented in Table 1. These numbers were generated using the linear trendline function in Excel.
Not so visually obvious are the declining slopes during the earlier portion of the data set. Table 2 presents the same characteristics for data chunks of approximately 20 year duration.
We saw in part 1 of this series that the relationship of consumption to net worth was not stable, so this result is not surprising. And we can now see that as net worth increases, the sensitivity of consumption to increasing net worth decreases.
I can think of two contributing factors. As wealth increases, the need to spend on basic necessities captures a smaller portion of that wealth, so the propensity to spend decreases. I’ll defer consideration of the other factor for now.
Here is a detailed look at how the slope of the PCE vs net worth line varied over time. Graph 2 shows the 34 quarter slope values for the data points of graph 1. The slope is plotted in dark blue, with certain time spans highlighted in contrasting colors: recessions are in orange, and the stock market and housing bubbles are in yellow.
Graph 2 Slope of Consumption vs Net Worth
Observations: 1) Except for the bubbles and the spike in the post-bubble recession of 2008, the values are mostly contained between a low of 0.168 and a high 0.246. 2) There was an upward trend that ended in the mid-70’s, underscored with a blue line. 3) Values after the mid-70’s, including the two bubbles, are contained in a down-sloping channel, outlined in green. 4) Except for the early 80’s and 90’s events, recessions are marked by sharp, temporary slope increases. 5) The average slope is 0.184, with a standard deviation of .038 6) The bubbles highlighted in yellow in Graph 2 correspond exactly to the data points in Graph 1 that fall below the red best fit line. 7) The post-bubble recessions brought the slope back into the range described in Observation 1. This is illustrated in Graph 1 by the returns to the blue best fit line.
If the normal relationship between net worth and consumption is described by a slope in the range of around 0.17 to 0.25, what is there about bubbles that causes drops into the range of 0.11 to 0.12 at the peaks? I think the answer is the second factor that I defered until now. The stock bubble and the housing bubble represented wealth increases that were not shared equally across the population. Specifically, as I pointed out earlier, these assets are mostly owned by the richest population segment, and growth in wealth has excessively favored the top 1% of the population. They have the least propensity to spend, and this tendency drives the PCE slope into the low range.
This FRED graph illustrates the point in a different way.
Graph 3 PCE, Net Worth and Disposable Income
There are four lines, Net Worth in green (divided by 5 to put it on the same scale); Disposable Income in purple, PCE in red, and Disposible Income multiplied by 0.931 in blue. Note that the last two overlap almost perfectly, as I also pointed out earlier (see link above.)
The conclusions I’m drawing are 1) Since the bubbles increased wealth in a highly skewed fashion, the relationship between average wealth and consumption broke down. 2) When the bubbles burst, the normal relationship between wealth and consumption reasserted itself. 3) The underlying cause is that during the bubbles the relationship between wealth and income broke down, and afterwards reasserted itself. 4) The relationship between disposable income and consumption is robust across time and most extraordinary financial events. 5) All the foregoing suggest that if wealth distribution were more even across the population (and thus more closely tied to disposable income,) then the relationship between wealth and consumption would be more robust.
That got me thinking again about the issue of whether consumption spending is determined by income or wealth. Specifically, if consumption is determined by wealth, there should be peaks in consumption corresponding to the dot-com and housing bubbles shown on Graph 1. However, as Graph 2 shows, there were no such peaks.
Graph 2 Personal Consumption Expenditures
I’ve argued already that, contrary to standard economic thought, consumption is directly determined by income. (Posted at RB and at AB.) One observation was that consumption, as a fraction of income, didn’t vary much over time, averaging 90.1% with a standard deviation of 2.1%.
I took a similar look at consumption and net worth, data from Fred. The next three graphs show personal consumption expenditures (PCE) as a decimal fraction of net worth (blue, left scale) along with net worth (NW) (red, right scale) over different time spans.
Graph 3A Expenditures/Net Worth and Net worth, 1959-79,
Graph 3A spans from 1959 – the beginning of the data set – to 1979. Net worth rises exponentially as the population grows. Adjusting for population growth does not change the shape of the net worth curve, so, in the aggregate, we were becoming richer during those years. Note that PCE/NW follows a generally similar, though far bumpier trajectory. As I pointed out in the prior post, the personal savings rate also increased during this period, so the average worker was able to both save and spend more.
Graph 3B Expenditures/Net Worth and Net worth, 1975-90
Graph 3B spans from 1975 to 1990. Net worth continues on its exponential track. But, after about 1979, PCE/NW drops, reversing the prior trend. By 1990, PCE/NW is no greater than it was in the early 1960’s. Meanwhile, the personal savings rate also dropped – to a range below that of the early 60’s.
Graph 3C Expenditures/Net Worth and Net worth, 1989-2011
Graph 3C spans from 1989 through October, 2011. The exponential growth of net worth falters before and during the two most recent recessions. After about 1994, PCE/NW is a roller coaster ride. Of particular interest is the exactly contrary motion at a detail level between NW and PCE/NW, after about 1998. During the housing bubble of mid-last decade, PCE/NW hit an all time low.
What narrative makes sense of these three graphs? Here’s my attempt.
Through the 60’s and 70’s, the standard of living was increasing, as incomes and net worth rose together. This allowed more discretionary spending, and therefore, the fraction of NW that was spent increased.
In the 80’s, aggregate net worth continued to rise, but consumption spending, quite dramatically, failed to keep pace. Lane Kenworthy has repeatedly pointed out that middle class income growth has decoupled from general economic growth as the upper income percentiles have captured an increasing slice of total income. As the wealthy grew wealthier and the middle class fell behind, the fraction of NW that was spent declined – exactly the opposite of what should happen if increasing wealth determined spending. But exactly what should happen if increased wealth is diverted to the already wealthy who have less of a propensity to consume.
During the 90’s, growth in median family income and GDP per capita were close to parallel (see graph at the Kenworthy link) so there was a lull in the decoupling. For most of that decade, PCE/NW was close to constant at 0.18-.19. But while spending was kept level, the personal savings rate continued to fall.
During the current century, median family income has flat-lined, while GDP/Capita has continued to increase. The decoupling has resumed and the wealth disparity has widened. During the two wealth bubbles, PCE/NW declined dramatically. When the bubbles burst and net worth declined, PCE/NW increased back into the 0.18-.19 range. Most strikingly, from about 1998 on, the two lines in graph 3C exhibit exactly contrary motion at a detail level.
There was a tight relationship between Net Worth and consumption through the 60’s and the 70’s, when earnings growth kept up with GDP and wealth disparity was slight by current standards.
This relationship broke down during the 80’s – though one could argue as early as the mid 70’s – as aggregate wealth and working class income decoupled.
Most recently, the relationship between NW and PCE/NW is inverse. The big swings in NW that the bubbles provided also demonstrated that consumption spending does not depend on net worth.
As I indicated in the earlier post linked above, consumption spending does depend on disposable income, throughout the entire post war period. A simple look at readily available data casts grave doubts on the idea that wealth, and not income, determines consumption spending.
For the longer perspective, here is the data of Graphs 3 A-C on a single graph.
Graph 4 Expenditures/Net Worth and Net worth, 1959-2011
In part 2, we’ll look at how spending and Net Worth correlate.
This is another look at the idea I put forth here, that – contra the standard economic idea that consumption depends on wealth – I believe that consumption depends on income. It’s worth stressing that wealth and income are not independent variables. Wealth is the accumulation of unspent income plus returns generated on that wealth over time. Is it proper to say that wealth is a stock, and income is a flow?
I believe the evidence very strongly indicates that consumption – also a flow – is tied tightly and directly to income. This does not mean that wealth cannot play a part in consumption decisions. People make all kinds of decisions about all kinds of things, for all kinds of reasons. But consumption decisions are constrained, and there is no reason why they can’t be constrained in more than one way.
I think the idea that consumption depends primarily on wealth is intuitively weak because consumption is aggregated over the population, while wealth is concentrated in a small segment of that population. A person with little or no wealth will spend the next dollar meeting some unsatisfied need, while the person with lots of wealth has the option of devoting it to rent-seeking or accumulation in an off-shore shelter. According to data now more than a decade old, the richest 1% of households owned 38% of all the wealth; the top 5% owned over half, and the top 20% owned over 80% of the wealth. The trend towards rising inequality started in the mid 70’s.
A couple of proxies for wealth are home and common stock ownership. Excluding home-ownership, the wealth concentration is even more extreme, with the top 1% owning 50% of the non-home wealth. It’s difficult to determine the actual amount of stock ownership in private hands. A number arrived at by elimination leaves 36% among households, non-profits, endowments and hedge funds. Therefore, realistically, the bottom 99% of individuals share about 18% of all stocks with those other institutions. At the bottom end, the lowest 20% have either no wealth, or negative net worth.
People at the low end live close to subsistence. People in the middle live pay check to pay check. For the vast majority of the population, the next marginal dollar has a high probability of being used as a consumption expense.
That is my narrative to support the idea that consumption must necessarily be strongly dependent on income. Now, let’s look at some data, through 2009, from the U.S. Census Bureau, Table 678. The first graph shows Disposible Income (green) and personal Consumption Expenditures (red) back to 1929.
A careful look suggests a narrative about this relationship. First, consider the depression years. From 1932 to ’34, consumption averaged 99% of disposable income. People had needs, and used their limited incomes to satisfy them, as best they could. Then, during WW II, with rationing and other constraints, saving was forced, and consumption was artificially low. Consumption reached an all-time low of 73.3% of Disposable Income in 1944. Since shortly after WW II, changes in Disposable Income and Consumption have been in virtual lock-step. I’ve put lines in a contrasting color connecting selected points in the Disposable Income curve, and dropped parallel lines for the same years onto the Consumption curve. Since 1951, very wiggle in Income corresponds to a wiggle in Consumption.
Here is a scattergram of the two subject variables, with a best-fit straight line provided by Excel.
As has been pointed out to me, correlation is not causation. But – when one can construct a rational narrative that explains the data, the two series display absolutely congruous motion over several decades, and R^2 is over 0.99, I’m willing to go out on a limb and say the burden of proof is on the denialists.
Here is a look at Consumption as a percentage of Disposable Income, since 1951.
I’ve expanded the Y-axis. In a view of the entire 0 to 100% scale, the post-1950 line barely wiggles. Over a span of 6 decades, Personal Consumption has averaged 90.1% of Disposable Income, with a standard deviation of 2.12%.
The data points, average, and an envelope one Std Dev above and below the average are all displayed on the graph. Despite having two clearly defined and opposite tending trends, this is still a well behaved data set, with 39 of 58 (67%) of the points within the envelope.
The two minima are in 1982 and 1984, and the bottom trend lines converge in 1982, so that is a reasonable time to define as the break point. This also suggests a narrative. During the post WW II golden age, typical wage earners moved incrementally above the subsistence level. This gave them the opportunity to save a little bit. Since 1982, as wages stagnated, it became necessary to devote a higher percentage to Consumption. Sure enough, savings grew through the mid 70’s, and have dropped dramatically since 1982 (or a bit earlier,) as this FRED graph demonstrates.
I won’t say that Consumption Spending is solely dependent on Income. But I will say that it is strongly, and even predominantly, dependent on income. Wealth might enter into the decision for those who actually have some, but they are in the minority and have few needs that can be satisfied by the next dollar of consumption.
My conclusion is that the best solution to the aggregate demand shortfall problem is to put money into the hands of the people who will actually spend it, and that the best way to do that is to give them jobs. As stop-gaps, various relief and welfare programs also have their place. This is the rational for fiscal stimulus. Federal spending programs provide real jobs for real people, and they will spend their earnings. Arguing about whether this is hole-filling or pump-priming strikes me as being just about as important as arguing about how many angels can dance on a pin head.
The prospects for domestic demand in the US are not bright. The labor market barely generates jobs and fiscal policy is a drag. Americans are consuming; but there’s unlikely sufficient nominal income growth to stabilize consumption expenditure growth at current levels.
We’ve seen years where consumption growth outpaced income growth; but those periods of consumption were financed through leverage build – with financial conditions tight, the possibility of financing consumption outside the labor market is deteriorating (see the Banking and Finance section of the latest Fed Beige Book, not encouraging).
Consumption growth cannot outpace income growth indefinitely. Unless we get a true policy kick (by fiscal policy, admittedly), the cyclical recovery could be a thing of the past.
That was nominal growth – in real terms and on a historical basis, the story is just as bad. Don’t let anybody tell you that real consumption growth. At 1.8% Y/Y in August is anything but miserable, especially given that its annual pace is 1 ppt below the long run average, 2.8% Y/Y. Long run real income growth is even worse at 2.5 ppt BELOW the long run average, 2.8% Y/Y.
Uh huh – yes, the US economy has definitely avoided the ‘recession scare’…right. As I see it, the problem with policy these days is not size nor level, rather complacency.
Private Investment peaks in 2006. Dating the start of the current recession from December of 2007 still strikes me as being six months late.
Consumption goes up fairly steadily from 1981, encompassing 8% more of total GDP—around a 12% increase over less than thirty years. Coincidentally, the Reagan Revolution moves taxes more heavily onto consumers at the same time. (Note that only about half of the appreciation in consumption is reflected in the change in Net Exports.)
Government spending declines starting around 1991, when the Real George (H.W.) Bush breaks his “no nude Texans” pledge. The era of Big Government remains over until George “Dad’s Rolodex Got Me Another Job I Can Screw Up” (W.) Bush desperately needed people to be employed:
Those are my top-of-the-head ones. What are yours?
Hi. For a project I’m working on, I need to find estimates of a private consumption multiplier. That is to say, the multiplier resulting from changes in private consumption rather than, say, from changes in gov’t spending or taxation. Any pointers? Please let me know in comments are by email (mike period kimel at gmail.com) Thanks.
Confidence is important, since consumer spending accounts for the lion’s-share of aggregate spending. Consumer confidence measures are highly correlated with the annual growth in real personal consumption expenditures – the correlation coefficients are 75% and 67% for the University of Michigan Sentiment index and the Conference Board’s Confidence index, respectively.
(Chart axis identifcation amended…rdan)
Ultimately, though, it’s all about jobs and personal incomes.
READ MORE AFTER THE JUMP!
To date, while July real wages and salaries (deflated using the CPI) fell on the month, the 3-month average continues its ascent. Clearly the sluggish climb in real wages and salaries is not enough to spark a surge in confidence and spending. Neither will consumers draw down saving, as was the case over the last decade amid debt-financed consumption. In fact, saving is more likely to rise as a share of income than fall as the balance sheet repair process furthers.
I’ll make this quick, since I’m going to get in trouble for writing on a national holiday. But the pace of annual jobs growth is too slow to generate strong wage and salary income. Much empirical research has been dedicated to the estimation of consumption functions, generally finding that labor income is the primary driver of consumption (here’s a primer at the Federal Reserve Board). However, by extension jobs growth is highly correlated with wage and salary growth, roughly 50% of personal income – this is the relationship I analyze here.
Roughly half of the BEA’s measure of personal income comes in the form of wage and salary, so called labor income and simply referred to as ‘wages’ from here on out. This is highly correlated with nonfarm payroll growth, both in nominal and real terms (92% and 79%, respectively, since 1996). The chart below illustrates the correlation between real wage growth and nonfarm payroll since 1982 (I use real wage so as to account for the effects of inflation).
The annual pace of real wage gains and jobs growth have declined over time (jobs growth is measured using the nonfarm payroll). Simply eyeballing the data, there’s a structural shift roughly around 1996, as listed in the table below.
Using these two time periods, 1982-1995 and 1996-05/2011, I estimate a simple model of real wage growth on nonfarm payroll growth. The chart for the 1996-2011 model is illustrated below; and for reference, the regression results across both time periods are copied at the end of this post.
READ MORE AFTER THE JUMP! Note: I do not have time for a full blown econometric analysis. I did, however, perform statistical tests for serial correlation in the errors, unit roots in the transformed data (none), and general modeling tests.
I come to two general conclusions regarding the relationship between real wage growth and jobs growth over time:
(1) Real wage growth has become more persistent over time. In the first period, 1982-1995, just one lag was required to expunge the errors of autocorrelation. Spanning the second period, 1996-2011, three lags were required. The sum of the coefficients on the three lags is 0.87 in the later sample, or current wage growth is highly dependent on previous periods – sticky if you will.
(2) Nonfarm payroll growth has become less significant over time. Spanning the years 1982-1995, the coefficient on annual payroll growth was 0.27 – for each 1pps increase in the annual payroll growth, the trajectory of annual real wage growth increased by 0.27pps. The coefficient dropped to 0.17 in the sample spanning 1996-2011. This is probably a consequence of service sector jobs growing as a share of the labor market. I’d like your ideas in comments as well.
Clearly this is a very simple model but it does highlight that wages are likely stuck in the mud for some time. In May, annual real wages fell 0.6% over the year, having decelerated for 5 of the 7 months since November 2010. Real wages can pick up, but it takes time AND jobs growth faster than the 0.67% annual pace in May 2011.
Ultimately, what this analysis tells me is with wealth effects slowing markedly – the trajectory of the S&P decelerated and house prices continue to fall – it’s going to take a burst of payroll growth to get real wage and salary growth back on track enough to finance US domestic consumption. One caveat to all of this negativity is that oil prices are coming off – this will boost real wage and salary growth directly.
Rebecca Wilder P.S. I guess this turned out somewhat less ‘quick’ than I had anticipated – not in trouble yet! Gotta go.
Readers of this blog know that I am in finance, specifically global fixed income. This blog post covers wealth effects in the financial industry, which is a relatively dominant share of total US compensation, 7.3% in 2009 and likely higher now (data are truncated at 2009). My view is that economists underestimate the wealth effects on consumption in the financial industry, given that financial wealth affects not only portfolio net worth but also the present value of labor income. Therefore, the sell-off in global risk assets may hit consumption more than expected in coming quarters, given that finance is the fifth largest industry, as measured by total compensation, on average spanning the years 1989-2009.
Why US consumption matters. The outlook for the US economy is of utmost importance to that for the world, given that the US will hold an average 22.1% of World GDP through 2016 (measured in $US), according to the IMF April 2011 World Economic Outlook. And the outlook for the US consumer is of utmost importance to that of the US economy, given that personal consumption expenditures hold a large 71% share of 2010 US GDP. Therefore, holding the US consumer share constant, US consumption is expected to be 15% of the global economy on average through 2016.
How wealth usually matters for US consumption. In economics, one of the drivers of consumption patterns ‘now’ is the wealth effect, usually defined as the shift in consumption due to changes in tangible (home values) and intangible (paper assets, like stocks and bonds) net assets.
(click to enlarge)
(Read more after the jump!) The chart above illustrates the ‘wealth effect’ on consumption as the ratio of net worth to disposable income (blue dotted line) as it’s correlated to the consumption share (outlays really, see table 1 for the breakdown) of disposable income (green line). The consumption (outlay) share is is 100 less the saving rate.
A large part of the Fed’s quantitative easing program (QE) was targeted at stimulating the positive wealth effects on consumption via higher risk asset prices. I would argue that this has been largely successful to date. The two year moving average of the consumption share (green solid line) fell precipitously following the financial crisis, only to generally stabilize since Q1 2009; this is largely coincident with the outset of QE1.
Back to why I brought up finance. There’s another effect in play here, more specifically related to the compensation structure in the financial industry. See, along with the tangible and intangible net asset values, total wealth includes the present value of labor income, i.e., the present value of lifetime compensation.
For all industries except finance, lifetime income is generally not associated with financial markets and risk assets, except via interest payments on fixed income. However, in finance total compensation is directly impacted by asset values via the bonus structure, often a large part of total compensation. Therefore, when asset markets are challenged, this likely affects the present-value of labor income adversely.
There’s an outsized wealth effect of net asset values in the financial industry: the direct wealth channel (net asset worth) plus the indirect channel (present value of labor income) on consumption.
Why is the financial industry important? It’s pretty simple: financial compensation is a large part of total US compensation, 7.3% in 2009, which has grown an average of 6% annually since 1988 in nominal terms. (Note: you can get this data from the BEA’s industry tables).
As financial markets take a turn for the worse – the S&P grew 5.4% December 31, 2010 through March 31, 2011 and is now down 4.1% since March 31, 2011 – the adverse wealth effect is likely to be stronger in the financial industry than in any other industry. For north of 7% of total US compensation, labor income is challenged in expectation, which is likely to drag consumption.
Purely anecdotal evidence. This strong wealth effect exists in my household. Both my husband (equities) and I (fixed income) are in finance; and when markets are challenged, we tend to save more. And it’s not because our stock portfolio is showing holes – actually, we don’t have much of a stock portfolio – it’s because our household income falls in expectation via the ‘bonus’ component of financial salaries.
I haven’t seen any work done on this wealth effect channel – but it does beg the question of whether there will be further downgrades to the US economic forecast if risk assets continue to sell off.