Friday, February 26, 2016

Humbling News for Macro-Finance Forecasters

The two graphs below, one from Deutsche Bank for the U.S. 10-year bond yield and one from the IMF for global real GDP growth, speak for themselves. Forecasting through large structural change is always difficult; maybe the means to which the forecasts should revert have now shifted lower. On the other hand, it's not obvious that we are experiencing a large structural change -- maybe we were just hit by a long and unlikely sequence of negative shocks, and the mean reversion embedded in the forecasts was ex ante rational even if ex post incorrect. 

Screen Shot 2016 02 24 at 7.05.40 AM

Sunday, February 14, 2016

David Hendry on Measurement, Theory, and Model Selection

David Hendry has an interesting new paper on the interplay of measurement and theory in model selection, "Deciding Between Alternative Approaches in Macroeconomics." As usual, David frames deep and important questions, and he furnishes penetrating insights as he works toward answers. Of course there's lots to disagree with, but that's beside the point -- original thinking is usually unsettling and controversial.

On David's theme of doing credible empirical work while maintaining focus on theory, see the No Hesitations post, "Shrinking VAR's Toward Theory." More generally, on his broad theme of the interplay between measurement and theory, see my The Known, the Unknown and the Unknowable (KuU) in Financial Risk Management, with Neil Doherty and Dick Herring. The introduction is freely available here. And don't forget another earlier No Hesitations post, "Theory gets too Much Respect, and Measurement Doesn't get Enough". (!)

Thursday, February 11, 2016

New R Code for High-Frequency Financial Data Analysis

I looked through the manual (below). Looks well done.

From the email:

Package features estimators for working with high frequency market data.

Microstructure Noise:
- Autocovariance Noise Variance
- Realized Noise Variance
- Unbiased Realized Noise Variance
- Noise-to-Signal Ratio

Price Variance:
- Two Series Realized Variance
- Multiple Series Realized Variance
- Modulated Realized Variance
- Jump Robust Modulated Realized Variance
- Uncertainty Zones Realized Variance
- Kernel Realized Variance (Bartlett, Cubic, 5th/6th/7th/8th-order, Epanichnikov, Parzen, Tukey-Hanning kernels)

Price Quarticity:
- Realized Quarticity
- Realized Quad-power Quarticity
- Realized Tri-power Quarticity
- Modulated Realized Quarticity

- for R (@ CRAN):

Tuesday, February 9, 2016

Worst Practices Conference

This ad just arrived in the email.  What a title.  Presumably the conference is about improving worst-case outcomes in order to improve expected minimax loss.  But still, that title...

2016 Foresight Practitioner Conference:
Worst Practices in Forecasting and Planning: 
Making Better Mistakes Tomorrow

Foresight has teamed up with the Advanced Analytics Institute at North Carolina State University (NCSU) in Raleigh to offer a tantalizing 1.5-day conference.

New Judea Pearl Causal Inference "Primer"

Should be a fun and informative read. Check out the contents and various chapters here. ("Causal Inference in Statistics - A Primer" by J. Pearl, M. Glymour and N. Jewell. Available now on Kindle; available in print Feb. 26, 2016.)

Saturday, February 6, 2016

Dual Regression

Speaking of quantiles and quantile regression, I also like the new version of Spady and Stouli's "Dual Regression." The power and insights of quantile regression, without the possibility of intersecting conditional quantile surfaces.  Sounds good to me.

Multivariate Quantiles

This new paper got me thinking. How often one wishes for a natural notion of multivariate median, or more generally, multivariate quantiles. Had fun learning about centerpoints and Tukey depths.

Multiple-Output Quantile Regression
By: Marc Hallin ; Miroslav Šiman

(Skip the first page, which is evidently from a different paper.)