This document discusses different approaches to measuring economic resilience. It notes that while resilience requires some redundancy which is inefficient, it also brings out an interesting trade-off between redundancy and efficiency. The document then evaluates the OECD's proposed measure of the "usefulness" of indicators for assessing resilience. It finds the OECD measure is similar but not identical to a more standard expected loss approach. It also discusses the OECD using quantile regressions to assess the impact of different factors on the tails of the GDP distribution, but notes concerns about identifying causal effects and justifying pooling across countries and periods.
HLEG thematic workshop on measuring economic, social and environmental resilience, Daniele Terlizzese
1. How to measure the resilience of
economic systems?
Comments by Daniele Terlizzese (EIEF)
Strategic Forum, 2015
Measuring Economic, Social and Environmental
Resilience
2. Schizophrenic comments
On the one hand, some skepticism on the concept of
economic resilience
On the other hand, some nitty-gritty details on the
empirical approach (OECD; too late for Philipp!)
Outline
3. Shocks happen. An efficient economy adapts in a way
that cannot be improved for everybody
Does a resilient economy do something else? Is
resilience an additional requirement?
If shocks are the โusual suspectโ (only big), resilience is
just a different name for efficient
More useful insights if we focus on โunusualโ shocks:
natural disasters
unforeseen contingencies
Is resilience a useful concept?
4. The economic system is not โdesignedโ to deal with
either kind of shocks, as they are:
either outside the realm of economics, or unamenable
to standard mechanism design
In both cases resilience seems to require idle capacity,
redundancy
But redundancy is ex-ante inefficient โ
resilience brings out an interesting trade-off
Redundancy vs. Efficiency
5. Excessive leverage, excessive maturity mismatches,
excessive debtโฆexcessive relative to what?
The analysis lacks a clear benchmark.
A (negative) shock would impose a cost even to a fully
efficient economy
How much larger is the cost actually borne?
Why is it larger? What are the market failures?
Distributive concerns? Need for a structural model
How to detect (lack of) resilience
6. OECD proposes a measure of โusefulnessโ of an indicator w
๐ข๐ข ๐ค๐ค = min(Pr ๐ถ๐ถ ๐๐, Pr ยฌ๐ถ๐ถ 1 โ ๐๐ ) โ
๐๐๐๐๐๐๐๐ ๐๐ ๐๐ ๐ค๐ค ๐๐๐๐ ๐๐๐๐๐๐ ๐ข๐ข๐ข๐ข๐ข๐ข๐ข๐ข
(๐๐Pr(๐ถ๐ถ) Pr(๐ค๐ค = 0|๐ถ๐ถ)
๐๐๐๐(๐ก๐ก๐ก๐ก๐ก๐ก๐ก๐ก ๐ผ๐ผ ๐๐๐๐๐๐๐๐๐๐)
+ 1 โ ฯ Pr(ยฌ๐ถ๐ถ) Pr ๐ค๐ค = 1 ยฌ๐ถ๐ถ
๐๐๐๐ ๐ก๐ก๐ก๐ก๐ก๐ก๐ก๐ก ๐ผ๐ผ๐ผ๐ผ ๐๐๐๐๐๐๐๐๐๐
๐๐๐๐๐๐๐๐ ๐๐ ๐๐ ๐ค๐ค ๐๐๐๐ ๐ข๐ข๐ข๐ข๐ข๐ข๐ข๐ข
)
Not standard (why should PM be concerned with T1 and T2
errors per se, rather than with actions?)
Compare with a more standard approach
Usefulness of indicators
7. Let
๐๐ 1 = ๐๐๐๐๐๐๐๐๐๐๐๐ ๐ฟ๐ฟ ๐๐ ๐ถ๐ถ Pr ๐ถ๐ถ ๐ค๐ค = 1 + ๐ฟ๐ฟ ๐๐ ยฌ๐ถ๐ถ Pr ยฌ๐ถ๐ถ ๐ค๐ค = 1
and
๐๐ 0 = ๐๐๐๐๐๐๐๐๐๐๐๐{๐ฟ๐ฟ ๐๐ ๐ถ๐ถ Pr ๐ถ๐ถ ๐ค๐ค = 0 + ๐ฟ๐ฟ ๐๐ ยฌ๐ถ๐ถ Pr ยฌ๐ถ๐ถ|๐ค๐ค = 0 }
The standard definition of the minimal expected loss from the
use of w is:
Pr ๐ค๐ค = 1 [๐ฟ๐ฟ ๐๐ 1 ๐ถ๐ถ Pr ๐ถ๐ถ ๐ค๐ค = 1 + ๐ฟ๐ฟ ๐๐ 1 ยฌ๐ถ๐ถ Pr ยฌ๐ถ๐ถ|๐ค๐ค = 1 ]
+
Pr ๐ค๐ค = 0 [๐ฟ๐ฟ ๐๐ 0 ๐ถ๐ถ Pr ๐ถ๐ถ ๐ค๐ค = 0 + ๐ฟ๐ฟ ๐๐ 0 ยฌ๐ถ๐ถ Pr ยฌ๐ถ๐ถ|๐ค๐ค = 0 ]
The loss from the use of w (1)
8. Assume/normalize ๐ฟ๐ฟ ๐๐(1) ๐ถ๐ถ = ๐ฟ๐ฟ ๐๐(0) ยฌ๐ถ๐ถ = 0 (with l.o.g.)
then the loss becomes:
Pr ๐ค๐ค = 1 [๐ฟ๐ฟ ๐๐ 1 ยฌ๐ถ๐ถ Pr ยฌ๐ถ๐ถ|๐ค๐ค = 1 ] +
Pr ๐ค๐ค = 0 [๐ฟ๐ฟ ๐๐ 0 ๐ถ๐ถ Pr ๐ถ๐ถ ๐ค๐ค = 0 ]
= ๐ฟ๐ฟ ๐๐ 1 ยฌ๐ถ๐ถ Pr ๐ค๐ค = 1, ยฌ๐ถ๐ถ + ๐ฟ๐ฟ ๐๐ 0 ๐ถ๐ถ Pr(๐ค๐ค = 0, ๐ถ๐ถ)
Compare this with the OECD formulation:
1 โ ฯ Pr(ยฌC)Pr ๐ค๐ค = 1 ยฌ๐ถ๐ถ + ๐๐ Pr ๐ถ๐ถ Pr ๐ค๐ค = 0 ๐ถ๐ถ
= 1 โ ฯ Pr(๐ค๐ค = 1, ยฌ๐ถ๐ถ) + ๐๐Pr(๐ค๐ค = 0, ๐ถ๐ถ)
Similar! But in general not true that
๐ฟ๐ฟ ๐๐ 1 ยฌ๐ถ๐ถ + ๐ฟ๐ฟ ๐๐ 0 ๐ถ๐ถ = 1
The loss from the use of w (2)
9. Let ๐๐๏ฟฝ = ๐๐๐๐๐๐๐๐๐๐๐๐ ๐ฟ๐ฟ ๐๐ ๐ถ๐ถ Pr(๐ถ๐ถ) + ๐ฟ๐ฟ ๐๐ ยฌ๐ถ๐ถ Pr(ยฌ๐ถ๐ถ)
Then the standard definition of the minimal expected loss
when w is not used is:
๐ฟ๐ฟ ๐๐๏ฟฝ ๐ถ๐ถ Pr(๐ถ๐ถ) + ๐ฟ๐ฟ ๐๐๏ฟฝ ยฌ๐ถ๐ถ Pr(ยฌ๐ถ๐ถ)
Suppose instead that the PM, when not using w, chooses
either ๐๐๐ถ๐ถ
or ๐๐ยฌ๐ถ๐ถ
, where
๐๐๐ถ๐ถ = ๐๐๐๐๐๐๐๐๐๐๐๐ ๐ฟ๐ฟ(๐๐|๐ถ๐ถ) and
๐๐ยฌ๐ถ๐ถ = ๐๐๐๐๐๐๐๐๐๐๐๐ ๐ฟ๐ฟ(๐๐|ยฌ๐ถ๐ถ).
The loss if w is not used
10. Then the minimal expected loss would be either
Pr C ๐ฟ๐ฟ ๐๐๐ถ๐ถ ๐ถ๐ถ + Pr(ยฌC) ๐ฟ๐ฟ ๐๐๐ถ๐ถ ยฌ๐ถ๐ถ or
Pr C ๐ฟ๐ฟ ๐๐ยฌ๐ถ๐ถ
๐ถ๐ถ + Pr(ยฌC) ๐ฟ๐ฟ ๐๐ยฌ๐ถ๐ถ
ยฌ๐ถ๐ถ
If we assume that ๐ฟ๐ฟ ๐๐๐ถ๐ถ
๐ถ๐ถ = ๐ฟ๐ฟ ๐๐ยฌ๐ถ๐ถ
ยฌ๐ถ๐ถ = 0 (with l.o.g.)
the minimal expected loss would be
min(Pr C ๐ฟ๐ฟ ๐๐ยฌ๐ถ๐ถ ๐ถ๐ถ , Pr(ยฌC) ๐ฟ๐ฟ ๐๐๐ถ๐ถ ยฌ๐ถ๐ถ )
This is similar to the OECD formulation:
min(Pr ๐ถ๐ถ ๐๐, Pr ยฌ๐ถ๐ถ 1 โ ๐๐ ). But:
(i) assumes a ยซbang-bangยป behavior of PM;
(ii) ๐ฟ๐ฟ ๐๐ยฌ๐ถ๐ถ ๐ถ๐ถ and ๐ฟ๐ฟ ๐๐๐ถ๐ถ ยฌ๐ถ๐ถ do not add up to 1;
(iii) ๐ฟ๐ฟ ๐๐ยฌ๐ถ๐ถ ๐ถ๐ถ โ ๐ฟ๐ฟ ๐๐ 0 ๐ถ๐ถ and ๐ฟ๐ฟ ๐๐๐ถ๐ถ ยฌ๐ถ๐ถ โ ๐ฟ๐ฟ ๐๐ 1 ยฌ๐ถ๐ถ
The loss if w is not used (2)
11. Recent strand in OECD analysis: assess the impact of
different factors on the tails of the GDP distribution
Quantile regressions are a useful and natural tool for
this, and are more robust to outliers
(but outliers might contain a lot of info on the tails)
Two concerns:
โข Correlation, not causal effect. How to identify the
latter?
โข Pooling across many countries and many periods
might not be justified
Quantile regressions