Zero return autocorrelation when broker is inactive

[sbenthall]

We are getting positive return autocorrelation with an inactive broker. That is not well calibrated!

[mesalas]

@sbenthall can you point me to a config that has positive autocorrelation ill pull the data and investigate whats going on.

[sbenthall]

'/shared/home/ammpssharkfin/master/allSharkfinrep10data/all/whiteShark10reps1001-whiteShark10reps_sim_stats.txt'

[sbenthall]

One explanation for this is that we are including the dividend in the 'return' for the risky asset, and the dividend growth (mean of the random walk) is positive.

[sbenthall]

A misunderstanding! The Durbin-Watson test is 2.0 with 0 autocorrelation, < 2.0 for correlation, and > 2.0 for anticorrelation. We have been subtracting 2. So a positive DW - 2 test means anticorrelation.

So:

• our log returns are anticorrelated because of the mean reversion effect of the institutional investors (though in reality this should be 0)
• they become correlated when there's a market failure (because the market is crashing)
• positive correlation of the log squared returns, which is volatility clustering, which is expected!
• more volatility clustering in market failure because things are rapidly getting worse.

[sbenthall]

A trend following agent in the market maybe could reduce the anticorrelation of returns by arbitraging on the institutional investor behavior.

[sbenthall]

@sbenthall will make the numbers produced by SHARKFin less confusing or better documented.