[R-SIG-Finance] portfolio optimization-autocorrelation in asset returns

Charles Ward cwrward at gmail.com
Wed Jul 23 14:30:03 CEST 2008


Autocorrelaton in returns often occurs in thin markets, especially in
real estate. One conventional response in real estate is to use a
filter to de-smooth the returns.
One such filter is De-smoothed R(t) = ( ObsR(t) - k*ObsR(t-1) )/ (1-k)
where k = either the coefficient of the autocorrelation or something
less (chosen subjectively) and ObsR = the observed or reported
returns.  The result is to increase the variance of the return series.
The underlying assumption is that the observed returns reported are a
weighted average of the "true" de-smoothed return and the previous
period's return. In some cases, the filter will increase the
correlation between the series and other market returns. With a highly
autocorrelated series, the de-smoothed series becomes implausibly
volatile so researchers have adapted the approach to modify the effect
by reducing k until the de-smoothed series looks "reasonable".

In a study for the UK Investment Property Forum, "Index Smoothing and
the Volatility of UK Commercial Property", March 2007, the authors
reported that in a survey of 13 of the largest real estate fund
managers in the UK, 9 adjusted the returns using this kind of filter.

Charles Ward

2008/7/21 Alexander Moreno <alexander.f.moreno at gmail.com>:
> Hi,
> I'm running a Markowitz Optimization using an EWMA correlation forecast and
> weekly data to find the minimum variance portfolios, updated every week, for
> a basket of currencies.  I'm finding that the performance is somewhat
> wanting, but when I remove currencies with significant positive
> autocorrelation over the sample, the performance over the same sample
> improves substantially (I know, my description is somewhat vague and this is
> also cheating).  However, I believe this is due to autocorrelation violating
> assumptions in the Markowitz Optimization framework, and I'm wondering if
> anyone could point me towards the best ways to get around this problem that
> don't involve looking at the autocorrelation over a sample and then removing
> the currency from the optimization for the same sample.
> Thanks,
> Alex
>        [[alternative HTML version deleted]]
> _______________________________________________
> R-SIG-Finance at stat.math.ethz.ch mailing list
> https://stat.ethz.ch/mailman/listinfo/r-sig-finance
> -- Subscriber-posting only.
> -- If you want to post, subscribe first.

Charles Ward
ICMA Centre - The Business School for Financial Markets
The University of Reading
Whiteknights Park
Reading RG6 6BA - UK

Tel: +44 (0)118 378 6292

Email: c.w.r.ward at reading.ac.uk

The information contained in this communication is confi...{{dropped:12}}

More information about the R-SIG-Finance mailing list