请教做shit桶的，trading signal-based strategy 如何calibrate? - 未名空间MITBBS历史存档

国际科技财经博客移民网络热点娱乐民生时事公众号

Redian新闻

>未名空间

>Stock

请教做shit桶的，trading signal-based strategy 如何calibrate?

请教做shit桶的，trading signal-based strategy 如何calibrate?# Stock

g*u2011-04-26 07:04

1 楼

就拿最简单的RSI，MACD为例子。简单起见，只考虑单一信号。
naive strategy可以两个threshold值。低于多少就买，高于多少就卖。就两个参数。
容易calibrate.
可能对买还问题不是太大，这个方法对卖效果不好。那么就增加点策略的灵活性，比
如divergence. 绝对数值与相对数值(divergence)考虑起来，参数也多点。
俺没有自己考虑过这个问题。但是感觉起码这个单信号策略要能够小范围适应一下，比
如，补偿一点lead-lag.

C*a2011-04-26 07:04

2 楼

Do you have a concrete example?

g*u2011-04-26 07:04

3 楼

A concrete example may be to get the largest return using RSI(14) signal on
SPY.
What would the best strategy look like?

【在 C****a 的大作中提到】

: Do you have a concrete example?

c*t2011-04-26 07:04

4 楼

单一信号 is too simple.
some time the place of stop and target is also important since it may
make or break the same trade.
Also some form of filtering is needed to filter out "bad" or "conflicting
signals". To in crease the odds of winning for the remaining ones.

g*u2011-04-26 07:04

5 楼

All valid pionts!
I was using the single signal as a toy problem. The question is on how a
good strategy centering around the signal should be structured.
Then, parameters can be calibrated with historical data.

【在 c**t 的大作中提到】

: 单一信号 is too simple.
: some time the place of stop and target is also important since it may
: make or break the same trade.
: Also some form of filtering is needed to filter out "bad" or "conflicting
: signals". To in crease the odds of winning for the remaining ones.

c*t2011-04-26 07:04

6 楼

I think you need to have a theory first.
then try to use TA indicator for your theory.
Not just trying all possible TA indicators.
In fact, the theory is there and everyone knows it.
Trade with the trend.
But there is a lot of detail that needs to be worked out.
How do you define trend.
How do you know you are not at the end of a trend.
what is most likely the start of a trend.
Where should be the stop so that it is likely the trend has changed.
etc...
TA indicators can be used to address each of these issues.
Then you will end up a system that you can use.

【在 g*****u 的大作中提到】

: All valid pionts!
: I was using the single signal as a toy problem. The question is on how a
: good strategy centering around the signal should be structured.
: Then, parameters can be calibrated with historical data.

w*s2011-04-26 07:04

7 楼

will read this thread after market close.

g*u2011-04-26 07:04

8 楼

I'll need more time to digest. Thank you.
My initial question remains. When confined to such a small technicality, how
to resolve it?
The motivation is a parameterized system where parameters of strategies can
be calibrated and performance of strategies can be compared.

【在 c**t 的大作中提到】

: I think you need to have a theory first.
: then try to use TA indicator for your theory.
: Not just trying all possible TA indicators.
: In fact, the theory is there and everyone knows it.
: Trade with the trend.
: But there is a lot of detail that needs to be worked out.
: How do you define trend.
: How do you know you are not at the end of a trend.
: what is most likely the start of a trend.
: Where should be the stop so that it is likely the trend has changed.