JCL has provided some useful info on prediction error in the past, but I have another question. In the performance report below, I get a large negative prediction error. The strategy creates a stationary spread by combining four JPY pairs and enters and exits positions in each of the pairs based on the spread's deviation from a recent mean value (loving the R bridge, by the way).

What does the large negative prediction error indicate? Would its negative value be related to the fact that even though the strategy is profitable, its average trade profit in pips is negative? I'm not sure exactly how performance error is calculated, so perhaps it isn't even applicable for spread trading or basket-type trading strategies??

BackTest MR_spd_4_grid portfolio

Simulated account AssetsFixFXP_Pers.dta
Bar period 1 hour
Test period 02.01.2008-24.07.2015
Lookback time 16 bars (16 hours)
Monte Carlo cycles 200
Assumed slippage 10.0 sec

Gross win/loss 163841$ / -155454$ (-22220p)
Average profit 1110$/year, 92$/month, 4.27$
Max drawdown -2060$ 25% (MAE -2060$ 25%)
Total down time 94% (TAE 97%)
Max down time 164 weeks from Feb 2012
Max open margin 228$
Max open risk 70318$
Trade volume 75811933$ (10032569$/year)
Transaction costs -7381$ spr, 178$ slp, -410$
Capital required 1526$

Number of trades 33387 (4419/year, 85/week, 8/day
Percent winning 50%
Max win/loss 309$ / -211$
Avg trade profit 0.25$ -0.7p (+26.0p / -24.7p
Avg trade slippage 0.01$ 0.0p (+0.3p / -0.3p)
Avg trade bars 9 (+8 / -8)
Max trade bars 94 (3 days)
Time in market 717%
Max open trades 31
Max loss streak 31 (uncorrelated 16)

Annual return 73%
Profit factor 1.05 (PRR 1.04)
Sharpe ratio 1.17
Kelly criterion 1.87
R2 coefficient 0.000
Ulcer index 8.8%
Prediction error -161%