US Homeowners Confound Predictions
The Financial Times is discussing attitudes in Last year’s model: stricken US homeowners confound predictions.

When Ray McDaniel, president of Moody’s, addressed a debate in Davos last week, the mood was so hostile that some speakers joked that he was brave to appear “without a bodyguard”.

“There has been a failure in some of the key assumptions which supported our analysis and modelling,” Mr McDaniel admits. “The information quality deteriorated in a way that was not appreciated by Moody’s or others.” Mortgage borrowers, in other words, did not behave as expected.

When American households have faced hard times in previous decades, they tended to default on unsecured loans such as credit cards and car loans first – and stopped paying their mortgage only as a last resort. However, in the last couple of years households have become delinquent on their mortgages much faster than trends in the wider economy might suggest.

More­over, consumers have stopped paying mortgages before they halt payments on their credit cards or automotive loans – turning the traditional delinquency pattern on its head. As a result, mortgage lenders have started to face losses at a much earlier stage than in the past.

In particular, it seems that mathematical models used to predict future default rates, based on past patterns of losses, have gone wrong because they did not adjust to reflect shifts in household behaviour. Or, to put it another way, financiers have been tripped up because they ignored one of the most basic rules of investment, which is usually found in product literature: the past is not always a guide to the future.

“There has been a failure in some of the key assumptions which supported our analysis and modelling,” Mr McDaniel admits. “The information quality deteriorated in a way that was not appreciated by Moody’s or others.” Mortgage borrowers, in other words, did not behave as expected.

Nevertheless, one thing is clear: the credit crunch will force many institutions to rethink their reliance on backward-looking models and perhaps put a greater emphasis on behavioural economics. “Simply extrapolating from the past into the future is not good enough,” says one US policymaker. Or as the beleaguered Mr McDaniel at Moody’s adds: “We [in the ratings industry] know we have got to retool our processes.”

amazing similarity. backward looking models trying to predict the future. incomplete models being shown to be useless, even dangerous
the above is from http://globaleconomicanalysis.blogspot.com/

regards.. NK

The Extremes Index is an attempt to show whether, over time, the US climate is becoming more “extreme”. It looks at data involving

1. maximum temperatures,
2. minimum temperatures,
3. drought/wetness,
4. “floods” (amount of rainfall from heavy rains)
5. dry spells/wet spells

Basically it looks at the period 1910-2007, calculates the values that constitute “extreme” (usually the 10% largest and 10% smallest of the population) over that almost-100 year period and then notes for each year how much of the US (area-wise) experienced those “extreme” conditions. Over the entire period the average, by definition, is 10%, but individual years vary from that. One looks for patterns in the time series.

I’m sure I’ve explained that poorly. The link gives other verbage to explain their approach.

Now, an initial point is what is meant by “becoming more extreme”. To me it’s a widening of the range. For example, an increase in the spread between the 10′th percentile and the 90% percentile would mean, to me, that the climate has become more extreme.

The definition used by the Extreme Index, however, is sensitive to shifts in the mean if the population is (more or less) normally distributed. At first glance that may not appear to be the case, because their methodology combines “much above” and “much below” for a measure and it might seem that a shift in the mean would simply cause more “much above” and equally less “much below”, thus offsetting any shift in the mean.

But, if the population is (more or less) normally distributed then then offset is only partial. Here is a normal distribution, with the mean and the “much above” and “much below” regions marked and it’s based on a population of 100 years. Now suppose that the mean shifts by 0.5 for the next twenty years. Those twenty years barely affect the overall population so that the values which define much-above and much-below barely change. However, those those twenty years with the mean shift have seen many of their temperature measurements slip into the “much above” category due to the shape of the normal distribution curve.

I’ve tried to illustrate those thoughts here .

Now, as time goes on the overall population mean will also shift and this effect will diminish. But if one is looking at say a thirty year period of mean shift out of a 100-year population then the mean-shift effect is quite noticeable. One just needs to understand that what’s being observed is the mean shift.

This has a curious effect. If the mean US temperature rises by 1C then that appears as an increase in the Extremes Index. If the mean remains there long enough to affect the base population and then then mean drops by 1C, back to the original temperature, then that second shift also shows up as an increase in the Extremes Index. Damned if it warms, damned if it cools.

My main point is that it’s important that a casual user of the Extremes Index understands what’s being shown. There’s no right or wrong in this, as I see it.

I have some more notes concerning recent aspects of the index, for later.

Meant to say convection here ?

Yes, sorry.

I’ve started looking for data which will allow me to construct a reasonable update to the GCRI (see #105), in case an online update is unavailable. I’m doing this element by element.

One of the five elements is diurnal (daily max to min) temperature range. In AGW hypotheses the diurnal range decreases, as greenhouse gases have a stronger warming impact on nighttime than daytime temperatures.

In my search for data I noticed this paper which covers global (not just US) trends. Here is a key excerpt, which shows the global diurnal trend (DTR, the blue line).

Remarkably, in a period of rising global temperature, a rise which is attributed to AGW, the global diurnal range is moving sideways.

I suspect, based on a coarse global map in the article, that the US was also trendless.

Not much support for AGW there. Maybe the other elements will provide the smoking guns.

5,000 Quatloos to anyone who can find the updated GCRI.

OK, I now offer 10,000 Quatloos to anyone who can locate the updated Greenhouse Climate Response Index.

For good measure I’ll toss in the keys to a ’98 Toyota Sienna minivan (slightly used)

If you are only worried about using mercury thermometers on mountain tops, I’m reasonably sure you are worrying about something the continuum assumption is just fine. I think the pressure only drops about a factor of 2 or 3 at the top of Kilamajaro. With respect to worrying about non-continuum effects for mercury thermometers with conventional dimensions, this drop in pressure is entirely unimportant.

If you’re worried about something else, the physical issue you are describing is real, and people do deal with this. (Based on conversations in “unthreaded”, I get the impression Larry sometimes does.)

Let’s take a mercury thermometer. For it to change, it has to take energy from the surrounding air to the mercury (via glass). Let’s say the thermometer is in darkness and the air moves past it, so sunlight photons are excluded. When the air is not dense, lacking the STP 6 x 10^23 particles per gram molecule, there are fewer interactions per unit time and the work available to heat the mercury is lowered. In the extreme case, if only one air molecule hit the thermometer each day, it would hardly change.

Does this get the point over? I don’t have the capacity to calculate a similar effect on mountaintops to see if it is trivial. Also, the effect would vary with wavelength according to the operating principle of the temperature transducer (glass bulb bolometer, thermocouple, etc).