Monday, 8 May 2017

Wealth inequality

Radio NZ's Emile Donovan asked me for comment on the latest Rashbrooke wealth inequality paper on Wednesday last week. I'm not sure whether my reply's been of use there, but I'll copy it below.

Paper summary:

Rashbrooke et al use several waves of SoFIE data to look at inequality in asset holdings and mobility across net asset quintiles. They then provide several cross-tabulations of the data splitting things up by ethnicity, gender, and age. They show that there is greater mobility across quintiles over longer periods than over shorter periods, and that there is less movement from the top and bottom quintiles than from the middle ones. They also demonstrate that the bottom decile has net debt rather than net assets, and that net debt is dominated heavily for that cohort by student loan debt.


The paper does not tell us much that is not already known about wealth holdings. Statistics New Zealand regularly releases wealth inequality data, though that annual series is less detailed than that which can be obtained in the (now rather dated) SoFIE data, and that data is regularly well-canvassed. Credit Suisse similarly puts together annual estimates of household net asset holdings. Little prior work is cited in the paper, and little context is provided that would help in assessing whether the levels of wealth inequality and wealth mobility are higher or lower than in prior periods, or higher or lower than international benchmarks. What is the ‘right’ level of mobility or inequality? Rashbrooke appears to come at it from a perspective that existing levels are too high, but doesn’t provide any benchmark for assessing what the right level is. I tend to come at it from a more process-oriented perspective which suggests there is no particular ‘right’ level but rather right processes: if the mechanisms for generating wealth are fair, then the outcome is fair, but if wealth is generated through cronyism then resulting outcomes are unfair regardless of the percentage of wealth held by any particular cohort. But in either case, knowing whether wealth inequality in NZ is high or low in international context would be helpful. And the same for mobility.

If we start looking to international benchmarks:
  1. Le, Gibson and Stillman found that inequality in household net worth in New Zealand is broadly similar to that in most other countries for which data is available.
  2. Credit Suisse data suggests that wealth inequality is very low compared to other countries, but I note that there is a broad range of countries that have basically the same wealth inequality as New Zealand – I’ve attached that bit from their report [Table 3-1 here]. That Credit Suisse report echoed findings from Davies et al, NBER working paper 15508 (2009) showing that wealth inequality in NZ was lower than all but 19 countries in a dataset of 150 countries.
  3. Stats NZ has had the wealth share of the top 10% in NZ as consistent with a 19-country OECD average, and the proportion owned by the top 1$ matching the OECD average.

More worryingly, parts of the analysis suggest that the authors have not fully come to grips with the data they’re presenting. At page 25, they note that the poorest decile has $1.8b in housing assets but $6.1 billion in mortgage debt. It doesn’t seem to have occurred to them that this is odd. It is odd because banks tend not to lend 339% of the value of a house to the poorest households. That is what is implied by owning $1.8b in housing but having $6.1b in housing debt. LVR restrictions alone mean they can’t lend more than 80% of the value of the house, never mind 300%. There is something wrong in that data series. And it’s not a particular secret either. I talked with StatsNZ about it when they released their latest round of wealth statistics last year, and I blogged on it, and I had an NBR column on it. One of the problems is that Stats data can be years out of date while its mortgage data is up-to-date: that means it’s easy to get mortgage debt reported well above housing assets in a rising market because the asset values are a lagged measure – but that can hardly be all of it because house prices have gone crazy, but not that crazy. We should be hesitant to draw conclusions from the series because of this problem alone, but the authors seem not to have even noticed that it’s a problem.

Further, the cross-tabulations don’t provide anywhere near the value that they could have. For example, it is well known that Pasifica and Maori communities are disproportionately younger than Europeans, and that older cohorts are far wealthier than younger cohorts. That means you need to age-standardise anything looking at ethnic differences so that you’re not confounding ethnicity effects with age effects. But, again, the authors seem not to have noticed that this is a problem. It’s bizarre. They go from showing the differences in wealth by age to the differences in age by ethnicity, and nothing seemed to click that the two might be related. A 30 second Google search gave me the StatsNZ page showing that median age (as of Census 2013) for Europeans is 41 years, but median age for Maori is 23.9 years and for Pacific peoples is 22.1 years.

Similarly, if the Maori and Pacifica groups are disproportionately much younger, they’re disproportionately not going to be moving out of the lowest wealth quintile because that doesn’t happen until you’re older. A better approach would have sorted by both ethnicity and age so that they’d be comparing all ethnic groups restricted to those aged, say, 30-35. Or 50-55. Pick a few and then show the differences by ethnicity within those age cohorts.

You’d similarly want age-correction on the mobility statistics. The typical life-cycle has people starting with net debt, then building wealth, then dis-saving during retirement. So you get mobility upwards until retirement, then mobility downwards as assets get consumed. Nothing that the paper puts up tells us how much of the mobility they find is natural age progression stuff. Again, you’d want to age-stratify the cohorts so people are measured in each wave against their position in the life-cycle changes. This likely drives some of their findings of bunching in the top and bottom quartiles, but it’s impossible to tell how much without going in and redoing it myself.

What else. The authors correctly note that student loan debt muddies things. They don’t explicitly state why. If you buy a house and have a 100% mortgage on it, then that’s a net zero contribution to wealth: asset matches debt. If you take out a student loan and have higher expected future earnings, all of the debt counts against you but the expected future earnings don’t. Trinh Le’s work, cited earlier, finds that those with university degrees are three times wealthier than those without university degrees.

Finally, if we’re thinking about international comparisons, countries like NZ will look more unequal than they really are as compared to countries with private pension systems. The wealth inequality stats would count retirement savings. A lot of retirement savings in NZ is done through the state and NZ Super. The claim that everyone has on NZ Super is a substantial asset, equally owned across everybody (albeit with adjustment for differences in life expectancy). Leaving that out makes wealth in NZ look less equal than it really is.


  1. Thank for this Eric. Some of those pitfalls are super useful to read about, as I was thinking of using the SOFIE data at some point.

    Do you know how it's possible that the data messes up the assets vs. liabilities for housing? I thought SOFIE was a survey, so who are these people massively over-reporting their debts? And why are different variables collected across different time periods? (Or is that part of their paper referring to a summary from Stats data, not SOFIE?)

    RE: international comparisons and retirement assets. I'm not sure that will make a big difference for, say, NZ-US comparisons. In the SCF, for example, 401K/IRA accounts are a really small proportion of total wealth. And a lot of authors writing about that survey explicitly leave these assets out of inequality calculations (sometimes they'll add them back in as income for retirees, which I think makes sense).

  2. I encountered the 'housing debt >> housing assets' problem in the latest Stats survey rather than SoFIE, and there it was (partially) that Stats was pulling in housing valuations that were old and merging in data on mortgage debt that was up to date , but there were also some problems around use of trusts where the kid from a richer family could be helping to pay the mortgage on a house he lives in that's held by the family trust and so his mortgage debt got recorded but the asset didn't - although that seems less likely to be a Decile 1 issue. In SoFIE, I don't know what the problem is. But there's no way that Decile 1 households, in the aggregate, have three times as much housing debt as they have housing assets.

  3. Ah ok. No, obviously that can't be the case. But I find it bizarre that that result could be coming from survey data. There's no way people can be *reporting* those kind of numbers, or anything close to them. It might be that they're imputing a lot of missing values (either on the asset or debt side), and something is going wrong with the conditioning information for Decile 1.