Tuesday, August 30, 2022

How Did I Miss This??

Great stuff, forthcoming JBES (2022).  



Abstract. The paper analyzes non-negative multivariate time series which we interpret as weighted networks. We introduce a model where each coordinate of the time series represents a given edge across time. The number of time periods is treated as large compared to the size of the network. The model specifies the temporal evolution of a weighted network that combines classical autoregression with non-negativity, a positive probability of vanishing, and peer effect interactions between weights assigned to edges in the process. The main results provide criteria for stationarity vs. explosiveness of the network evolution process and techniques for estimation of the parameters of the model and for prediction of its future values.


See also https://annabykhovskaya.com

Wednesday, August 24, 2022

The Complexity Principle (!)

Continuing the previous post, I'm sorry if I seem to be gushing over the recent Kelly et al. program (indeed I am), but it just blows me away.  The famous "parsimony" and "KISS (keep it sophisticatedly simple)" principles turned on their heads!  George Box and Arnold Zellner must be rolling in their graves...

 The Virtue of Complexity Everywhere

Bryan T. Kelly (Yale SOM; AQR Capital Management, LLC; National Bureau of Economic Research (NBER)); Semyon Malamud (Ecole Polytechnique Federale de Lausanne; Centre for Economic Policy Research (CEPR); Swiss Finance Institute); Kangying Zhou (Yale School of Management)

We investigate the performance of non-linear return prediction models in the high complexity regime, i.e., when the number of model parameters exceeds the number of observations. We document a "virtue of complexity" in all asset classes that we study (US equities, international equities, bonds, commodities, currencies, and interest rates). Specifically, return prediction R2 and optimal portfolio Sharpe ratio generally increase with model parameterization for every asset class. The virtue of complexity is present even in extremely data-scarce environments, e.g., for predictive models with less than twenty observations and tens of thousands of predictors. The empirical association between model complexity and out-of-sample model performance exhibits a striking consistency with theoretical predictions.


Friday, August 19, 2022

Complexity in Prediction

Really glad to see that Kelly et al. are keeping at it, moving well into the "double dip" zone and adding regularization.

 The Virtue of Complexity in Return Prediction (2022)

Bryan T. KellySemyon MalamudKangying Zhou

The extant literature predicts market returns with “simple” models that use only a few parameters. Contrary to conventional wisdom, we theoretically prove that simple models severely understate return predictability compared to “complex” models in which the number of parameters exceeds the number of observations. We empirically document the virtue of complexity in US equity market return prediction. Our findings establish the rationale for modeling expected returns through machine learning.


Wednesday, August 10, 2022

Instrumental Variables in Practical Application

I have always been fascinated by Alwyn Young's paper,  "Consistency without Inference:  Instrumental Variables in Practical Application."   On-line appendix.  Glad to see that it's now published in the European Economic Review.  Note the key role of non-white disturbances.

From the intro:

The economics profession is in the midst of a “credibility revolution” (Angrist and Pischke 2010) in which careful research design has become firmly established as a necessary characteristic of applied work.  A key element in this revolution has been the use of instruments to identify causal effects free of the potential biases carried by endogenous ordinary least squares regressors.  The growing emphasis on research design has not gone hand in hand, however, with equal demands on the quality of inference.  Despite the widespread use of Eicker (1963)-Hinkley (1977)-White (1980) heteroskedasticity robust covariance estimates and their clustered extensions, the implications of non-iid error processes for the quality of inference, and their interaction in this regard with regression and research design, has not received the attention it deserves.  Heteroskedastic and correlated errors in highly leveraged regressions produce test statistics whose dispersion is typically much greater than believed, exaggerating the statistical significance of both 1st and 2nd stage tests, while lowering power to detect meaningful alternatives.  Furthermore, the bias of 2SLS relative to OLS rises as predicted second stage values are increasingly determined by the realization of a few errors, thereby eliminating much of the benefit of IV.  This paper shows that these problems exist in a substantial fraction of published work.