I’ve been a bit slow to get around to writing about the material the Reserve Bank released last week about the dairy stress test it conducted with the five largest dairy-sector lenders late last year.

I’ve long been of the view (and on record here) that, almost no matter how severe the dairy situation becomes, dairy loans would not represent a threat to the soundness of the New Zealand financial system.  That is a top-down analysis based on

• the size of capital of the New Zealand banking system (around $36bn),
• the overseas ownership of all the main dairy-lending banks, and the absence of correlated exposures in most of them (dairy loans aren’t a big part of the Australian parents’ books),
• the fact that any losses on the dairy book will crystallise gradually, allowing other retained earnings, or outside injections of new capital, to buttress the overall position of the New Zealand banks, and
• that while dairy losses could in time be a part of a wider set of banking system losses (eg if severe losses also mounted on the housing portfolio), it is almost inconceivable that in such a scenario New Zealand’s exchange rate would not fall a lot further.  A NZD/USD exchange rate of, say, .39 (where it got to in 2000) covers over quite a lot of weakness in the international prices of whole milk powder (in turn mitigating the severity of the dairy losses themselves).

There are counters to each of these points, but in the end I think they come down to this: if the Australasian banks ever face really large losses on their housing loans, the banks could be in trouble.  I think that is very unlikely: house prices are held up by a combination of regulatory land use restrictions and population pressures, and vanilla housing lending has rarely if ever collapsed a banking system (as the Reserve Bank itself has acknowledged).  You might disagree, but my real point is that dairy loans themselves aren’t going to threaten the soundness of the system. Much wealth will be lost.  And many of the individual loans may have been ill-judged (by borrower and lender) but that is a different issue, and almost in the nature of a market economy operating under (the real world) conditions of uncertainty.

That is all top-down perspectives. But the stress test was useful precisely because it aims to be a bottom-up approach: working with the banks on how their actual dairy portfolios would behave under two pre-specified scenarios.  Note what the exercise wasn’t: it didn’t look at the implications for loans to dairy companies themselves, or to suppliers to the dairy industry (companies or farmer), and also didn’t look at the impact on loan losses elsewhere in the portfolio resulting from the stresses on the dairy sector itself (eg retailers or builders or residential mortgages or… in dairy-dependent towns).  Note that it was also only a rather provisional exercise, indicative more than definitive, and a basis for ongoing discussions between the Reserve Bank and individual banks.

That does tend to suggest we should use the higher loss estimates rather than the lower ones (since banks have fewer incentives to overstate the loss implications than to understate them).

Here are the scenarios the Reserve Bank specified.

I’m largely going to ignore Scenario 1 from here on.  As the long-term average real milk price is probably only around the assumed 2017/18 level, Scenario 1 doesn’t represent much of a stress test at all.  The banks and the industry would have to be have been very rickety for a scenario like that to have presented a banking system problem.  I think the Reserve Bank should also have discounted these results, rather than highlighting them in their press release.

Scenario 2 does look much more like a real stress-test.  But even if one thought the series of payout assumptions might be reasonable (2015/16 won’t have been that low, but some of the out years could still be lower than assumed here), I was surprised by the dairy land price assumptions.  Despite a really severe adjustment in the payout path (absolutely, and probably relative to farmer expectations), dairy land prices are assumed to fall by just under 40 per cent (the cumulative effect of those three annual falls).

That might sound like a lot, but:

• when the Reserve Bank did its housing stress test, it assumed a 50 per cent fall in Auckland house prices.  People still need to live somewhere, while they don’t need to farm cows.
• we’ve already a dairy land price scare not long ago.  Here is a chart of the (“hedonic”)dairy land price index the Reserve Bank developed for REINZ (despite which, we don’t have general access to the series).

In a single year, dairy land prices fell by more than 30 per cent –  and that was a severe, but very short-lived, fall in milk prices, and a rise in dairy non-performing loans that was still moderate compared to what we see in Scenario 2 in the current stress test.   Perhaps deliberately, the Reserve Bank’s stress test does not seem to have taken account of a second round of selling (forced or voluntary), and the potential for that to drive land prices well below what might be a longer-term equilibrium level.  Overshoots routinely happen in such markets, where liquidity is thin to non-existent, uncertainty is rampant, and potential buyers are few.  As Eric Crampton’s discussion highlights, one difference between now and 2009 will be that potential buyers are probably much more aware of how significant the barriers are to any offshore buyers (who might otherwise be a stabilizing force in the market).

Loan losses evaluated on total dairy land prices falls of perhaps 60 per cent might be a more realistic stress test –  recall, that stress tests aren’t central predictions, they are a scenario to test robustness against.  Loan losses went up by 5 percentage points on the move from the (not very stressful) Scenario 1 to Scenario 2.   The pattern of losses on loans should rise non-linearly as the test gets more stressful, and moving from a 40 per cent land price fall scenario to a 60 per cent scenario is a bit more of a land price adjustment than moving from Scenarios 1 to 2.

There are lots of other points of detail I could question (some things in the article just aren’t made as clear as they could be), but will just highlight one.

The Reserve Bank has long emphasized the desirability of having a capital framework for banks in which risk weights (whether imposed by the Bank, or flowing from the internal models of the major banks) do not have the effect of making capital requirements pro-cyclical.  If capital requirements fall in asset booms and rise in shakeouts, the capital requirements will tend to amplify credit and asset price cycles (an existing stock of capital will go ever further as the boom proceeds, and ever less far – encouraging banks to rein in lending even more –  as the bust proceeds).  And yet the stress-testing article suggests that pro-cyclicality is deeply embedded in the modelling, at least for the dairy portfolio –  itself the largest single chunk of banks’ commercial lending.

Here is what I mean.

This chart shows the average risk weight for the banks’ dairy portfolios under Scenario 1.  Recall that Scenario 1 was not very demanding at all, and yet the average risk weight on dairy loans increases by 60 per cent  (eg, from, say, 70 per cent to 112 per cent).  No doubt deliberately, the Bank does not reveal how much further risk weights increase in the much more onerous Scenario 2.     Even if it is not that much further, this sort of highly pro-cyclical pattern of risk weights looks like a bug that needs some serious attention.

To recap, in Scenario 2, bad debt expenses average 8 per cent of dairy exposures.

But, as the Bank noted (see extract above), not all banks were as conservative as others.  If we take the pessimistic end of the Scenario 2 range, we would have bad debt expenses of perhaps 11.5 per cent of dairy exposures.  But, as noted above, the near-40 per cent fall in land prices  in Scenario 2 still looks too shallow for such a fully-worked-through scenario.  If land prices were to fall 60 per cent would it be implausible that in such an scenario, with all the second round effects accounted for and allowing for the non-linear loss profiles, the banks could face losses not of the “3 to 8 per cent of their total dairy exposures”  that the Reserve Bank highlighted, but something more like 20 to 25 per cent of their total loans to dairy farmers?    I deliberately pose it as a question, rather than a confident assertion, and it is –  deliberately – the result of a stressed scenario, but it is probably a question people should be posing to the Reserve Bank.