I have a rotational trading system that gets creative in the position
sizing. I was wondering if the following is possible in Amibroker:
1. Rank a (small) universe of stocks by a single factor
2. Select the top x% and bottom x% to go long and short, respectively
3. Weight the individual long and short positions inversely
proportionally to their n-day volatility. For example, let's say the
system wants to go long and short two stocks on each side. Let IVL1 be
the n-day inverse volatility of long stock 1, IVL2, IVS1, and IVS2 be
the n-day inverse volatilities of long stock 2, short stock 1, and short
stock 2, respectively. Further, let WL1, WL2, WS1, and WS2 be the
percent weights of long stocks 1 and 2 and short stocks 1 and 2,
respectively. Then, the system wants WL1 = IVL1/(IVL1 + IVL2); WL2 =
IVL2/(IVL1 + IVL2); WS1 = IVS1/(IVS1 + IVS2); WS2 = IVS2/(IVS1 + IVS2).
4. Then, as a further refinement, the system wants to beta-adjust the
short positions relative to the long positions. That is, if the
portfolio of long stocks (weighted according to their volatilities) have
an m-day beta of B with the portfolio of short stocks (again, weighted
according to their volatilities), the system wants to create a
long/short portfolio that has B dollars short for every dollar long.
5. The system wants to buy on the close
6. The system will exit at the following day's open if the portfolio
is up at the open, otherwise it will to hold until the following day's
close, when it will repeat steps 1 through 5
I'd like to both backtest the system (I've backtested it previously in
Excel/VBA) and run it in real-time in Amibroker. I'd be grateful for any
insight as to whether the weighting scheme (both the
inverse-volatility-based weighting, and the beta adjustment) is doable
in Amibroker, and if so, how complex a program I'd be looking at. The
rest of the system seems relatively straightforward.
In case steps 3 and 4 above were not clear, here's a little more detail.
After finding WL1, WL2, WS1, and WS2, I'd like to form two time series
of historical returns. The first time series take the historical returns
of the two long stocks and combines them using WL1 and WL2 (i.e., long
portfolio, PL = WL1*R1 + WL2*R2, where R1 and R2 are the time series of
returns for long stock 1 and 2, respectively). The second does the same
for the short stocks using WS1 and WS2. Then given the two portfolios,
PL (long) and PS (short), I want to find the beta of PL to PS. Finally,
I want to have the ultimate portfolio, P, to be as follows: P = PL -
B*PS (where B is the beta of PL to PS). The final weights with the beta
included would be:
WL1 = IVL1/(IVL1 + IVL2)
WL2 = IVL2/(IVL1 + IVL2)
WS1 = B*IVS1/(IVS1 + IVS2)
WS2 = B*IVS2/(IVS1 + IVS2)
(But, of course, one needs to find the weights sans B first in order to
find B.) Would be grateful for any pointers as to how to go about
implementing this, if it's indeed possible.
Regards,
Ray
