问题
This is my final try after posting several questions that might have been badly addressed/described. I want to achive this DataFrame as result:
Signal price buy_units sell_units cashflow balance
Index
0 0 40 0 0 0 100000
1 1 50 2000 0 -100000 0
2 -1 100 0 -2000 +200000 200000
3 1 50 4000 0 -200000 0
4 -1 70 0 -4000 +280000 280000
It's a stock trading strategy backtest. When Signal is ==1 buy_units is equal to current balance(value from prior row) divided by price. balance then becomes reduced by cashflow (that is units x price*-1). The remainder should be self-explanatory.
The problem: calculating buy_units without iteration but in vectorized fashion. Who has an idea how to solve this?
ultimately I would like to design this in a class with a "Calculate units" and "calculate balance" method but this is not necessary in first step.
回答1:
If -1 and +1 signals always alternate as in the example, then the balance in step n+2 is equal to the balance in step n multiplied by the price return between n+1 and n+2.
I use the cumulated product to translate this into vectorized operations in pandas:
# initialize balance
df['balance'] = 0.0
df.balance.iloc[0] = 10000.0
# calculate returns
df['return'] = df.price / df.price.shift()
# calculate balance where signal is -1
df.loc[df.Signal == -1, 'balance'] = \
df.balance.iloc[0] * df.loc[df.Signal == -1, 'return'].cumprod()
The cashflows and number of units can easily be computed from the balance.
回答2:
You can create a temporary dataframe in order to do your computation:
temp = df[df.balance == 1]
Then perform computations on this dataframe:
df.ix[df.balance == 1,'buy_units'] = temp.balance/temp.price
And delete the temporary dataframe:
%reset selective -f temp
来源：https://stackoverflow.com/questions/37653822/vectorized-backtest-creation-of-pandas-dataframe