Portfolio optimization with “ Threshold Accepting ” : a practical guide 221
Throughout , we always assumed that we have a model, that we have the
data (i.e., scenarios), so that we could concentrate on the optimization proce-
dure. But setting up an optimization algorithm can only be the first step. The
main problem in portfolio optimization, no matter what technique is used, is
how to tell noise from information, i.e., how to reduce overfitting. (An unfor-
tunate rule of thumb in financial optimization: if something seems too good to
be true, then it is not true.) To improve in this direction, more effort needs to
be put into data modeling (i.e., the scenario generation process) and toward
the testing of the empirical effectiveness of different portfolio selection criteria.
Both issues are, in our view, strongly “ under-researched ” in the academic
world, at least when it comes to actual empirical performance. Note that this
strongly relates to the first two of our initial questions: “ what do I want to
optimize? ” and “ what can I estimate? ” Much research in portfolio optimiza-
tion relates to in-sample properties of different methods: given a data set, we
can now minimize drawdown, or ratios of losses to gains. But what we actually
want is to minimize future drawdown, or the ratio of future losses to future
gains. There exists comparatively little academic research into how these objec-
tives relate to the quantities that we actually optimize.
The advantage of heuristic optimization methods in this respect is that when
we formulate a model, we are quite unconstrained with regard to its tractabil-
ity, so the third question ( “ what can I compute? ” ) becomes much less of an
obstacle to progress. Optimization is a tool, and like with any tool, it is its
application that matters.
Acknowledgment
Both authors gratefully acknowledge financial support from the EU Commission
through MRTN-CT-2006-034270 COMISEF.
References
Alth ö fer , I. , & Koschnick , K.-U. ( 1991 ). On the convergence of “ threshold accepting. ”
Applied Mathematics and Optimization , 24 ( 1 ) , 183 – 195.
Bacon , C. A. ( 2008 ). Practical portfolio performance measurement and attribution ( 2nd
ed. ). New York: John Wiley & Sons.
Barr , R. S. , Golden , B. L. , Kelly , J. P. , Resende , M. G. C. , & Stewart , W. R. ( 1995 ).
Designing and reporting on computational experiments with heuristic methods.
Journal of Heuristics , 1 ( 1 ) , 9 – 32.
Biglova , A. , Ortobelli , S. , Rachev , S. , & Stoyanov , S. ( 2004 ). Different approaches to risk
estimation in portfolio theory. Journal of Portfolio Management , 31 ( 1 ) , 103 – 112.
Chekhlov , A. , Uryasev , S. , & Zabarankin , M. ( 2005 ). Drawdown measure in portfolio
optimization. International Journal of Theoretical and Applied Finance , 8 ( 1 ) , 13 – 58.
Dueck , G. , & Scheuer , T. ( 1990 ). Threshold accepting. A general purpose optimiza-
tion algorithm superior to simulated annealing. Journal of Computational Physics ,
90 ( 1 ) , 161 – 175.