Linda Mhalla, Julien Hambuckers, and Marie Lambert, "Extremal
Connectedness of Hedge Funds", Journal of Applied Econometrics,
Vol. 37, No. 5, 2022, pp. 988-1009.

The data used in the empirical analyses have been provided by Hedge Funds
Research (HFR) and are available at https://www.hfr.com/hfr-database
under proper licensing and registration.

All files are ASCII files in DOS format. They are zipped in the file 
mhl-files.zip. There is also a supplement in 
Mhalla-Hambuckers-Lambert-supplement.pdf.


Marginal analysis

Fund-specific variables have been provided by HFR. It includes the returns
(i.e. our target variable, taken as the adjusted performance calculated by
HFR), the asset under management (scaled by a factor of 1000), the level
of incentive and management fees, and the existence of a leverage or not.
HFR provided us also with the sub-strategy classification between the
12 investment styles considered in the the paper, as well as the country
of registration. A detailed description of which funds are excluded from
this initial sample (e.g. because of missing data, early inception date,
etc.) is given in Section 1 of the Supplementary material available at
https://github.com/lindamhalla/Extremal-connectedness-of-hedge-funds.

To associate macroeconomic indicators to each fund, we use the country of
registration. Macroeconomic variables (i.e. unemployment rate, short- and
long-term interest rates, industrial production) have been obtained from
the OECD website (see http://stats.oecd.org - where detailed definitions
of the time series are provided). Stock volatility has been calculated
as monthly average of 360-day volatility of the national stock market
index. Data for these calculations have been obtained from Bloomberg.

Finally, we use the publicly available 7 risk factors of David Hsieh
available at https://faculty.fuqua.duke.edu/~dah7/HFData.htm,
and the 5 factors available on Kenneth French website at
https://mba.tuck.dartmouth.edu/pages/faculty/ken.french/data_library.html

Time series of the macro-financial variables used in this paper are
made available in the file 'macro_finance.xlsx', Funds returns for
which macro-financial data are missing are excluded (this concerns
mostly small countries and a limited number of funds, since overall 95%
of our observations are tied to funds registered in the US, the UK,
Switzerland, France, and Canada, for which no macro-financial data are
missing). Factors data can be found in the files 'fung_hsieh.xlsx' and
'fama_french.xlsx'.

Remark that all returns have been multiplied by -1 and we focused on
the right tail, since our interest is on losses.

The code in 'script_marginal_analysis.m' allows estimating an extreme
value regression model, either with a likelihood or a penalized-likelihood
approach, as in the paper.


Multivariate analysis

After estimating the marginal tail distributions of the different
investment styles of the hedge funds, the negative returns are transformed
to the unit-Frechet scale using eq. (10), as described at the beginning
of Section 3.3 of the paper.

Then, (bivariate) vectors of transformed returns are built for each pair
of investment styles. Finally, only observations with a radial component
exceeding a high threshold are kept for inference, where a Husler--Reiss
spectral density is fitted.

In the text file 'df.multivariate.txt', which is a CSV file with spaces 
for separators, we provide the negative returns transformed to the unit 
Frechet scale (for anonymity reasons), as well as the set of covariates 
used to model the dynamic tail connectedness measures, i.e., the VIX 
(retrieved from the Federal Reserve of Economic Data), the negative 
returns of the MSCI (retrieved from the MSCI World page), the EPU 
(retrieved from the Federal Reserve of Economic Data), and the FSI 
(retrieved from the Federal Reserve of Economic Data).

The R script used to estimate the extremal connectedness between
investment styles of hedge funds (see Section 3.3 of the paper) can
be found in the file 'modeling_extremal_connectedness_funds.R'. This
R script allows also the computations of the ECoVaR for all pairs of
strategies (see Section 3.4 of the paper). The functions needed to fit
the Husler--Reiss spectral density can be found in 'spectral_HR.R'.