Clifford Asness is a hedge fund
manager and co-founder of AQR Capital Management, with degrees in English,
Economics and Finance. Together with his
colleague Aaron Brown, he recently published a note that analyzes global warming data and concludes that the
data do not support the projections made by climate-science based IPCC models. In these comments, I will point out some of
the problems with the analysis of Asness and Brown (hereafter referred to as
AB), which make their conclusions somewhat unsubstantiated and perhaps even
misleading.
AB start with the global land-ocean annual mean surface
temperatures, as shown in their Figure 1.
This is their “data”, where time is the independent variable or
“predictor”, and the global temperature index (i.e., deviation of the recorded
temperature from the 1951-1980 average) is the dependent variable or
“response”. They proceed to fit a linear
regression trend line to the data (Figure 2), and compare the projections from
this simple “model” with IPCC model results (Figures 3-5). Since the linear fit in Figure 2 does not
fully match the curvilinear trend in the data, especially for the post 1950
period, AB then fit a quadratic regression model to the data and compare the
corresponding updated projections to IPCC model results (Figure 6). Their main conclusion is that the global
warming trend, extrapolated from a time series of past temperature readings, is
at odds with IPCC projections. To quote
AB, “You can believe the models if you like, or you can look at the data
and assume the most likely future is an extrapolation of the past (my bold). What you cannot do is both.”
Let us consider the following thought experiment. I want to predict the future movement of the
S&P 500 index. Using AB’s analytical
premise, I should be able to look at the past history of this index and project
it into the future. To illustrate this
point, I will use a time series of the S&P 500 Index values from 1950 until
the present. The figure below shows a
chart of this data, where the index values are plotted in a logarithmic scale
per normal practice. Next, I fit a
log-linear model to this data, with an R^2 value of 0.9533, indicating an
excellent fit.
Even though this fit is as good as what AB show with their analysis,
does this give me any confidence that I can predict the future trajectory of
the S&P 500 based on past history? Not
really. Would any investment manager
make decisions based on these projections of the S&P 500’s values – without
doing his or her due diligence about future conditions? I think not. A reasonable analysis would require us to
consider how future estimates of corporate earnings, interest rates, Eurozone
conditions, investor sentiment, etc., compare to the immediate past and then
decide if inferences about the future can and should be made from past data.
Extending this thought process to AB’s analysis of the
global warming data, it is clear that time is simply a proxy for some
underlying causal variable. We can
assume that the future can be extrapolated from the past, if and only if
the time-dependence of key causal variables affecting global warming in the
past is similar to that expected for the future. In my view, this is the fundamental flaw in
AB’s analysis. The IPCC models have
various scenarios that describe how CO2 emissions and other drivers of global
warming can change with time as compared to current conditions. This is what is missing in the AB
analysis. Not only that, but AB also
appear to be unaware of the fact that all physics-based IPCC models are
calibrated to past temperature trends and are consistent with the data. The divergence in their projections is caused
by different assumptions about future conditions with respect to CO2 emissions. A credible model has to explain the past
before it can be applied for predicting the future.
In summary, as the simple S&P 500 index example is
intended to show, past is prologue only if “before” and “after” operating
conditions are similar. By most
accounts, that is not the case with our understanding of how climate is
changing. The
“climate-knowledge-free-statistics” based simplistic AB analysis fails to
properly take into account the influence and time-dependent behavior of the
relevant causal variables, thus leading to results that are incorrect and
misleading. If only AB had applied the
same analytical rigor to this problem as for an investment decision!
POSTSCRIPT: AB’s
central premise (i.e., the future can be extrapolated from the past) also
brings to mind the simplest weather prediction model – which posits that the
weather tomorrow will be the same as that today. Suffice it to say, in the words of the
inimitable H.A.L. Mencken, “for every complex problem there is a solution
which is simple, neat and wrong”.
Caveat Emptor.