j***y
发帖数: 41 | 1 怎么比较2个不同的models哪个好关于policy loss，他们从同一数据来，不同的
predictor. 不知道问题是不是这样，不是听的很清楚。 | G*****u
发帖数: 1222 | 2 monte carlo下 看看volatility的大小？
或者直接比较跟实际数据的偏差？ | G*****u
发帖数: 1222 | 3 exam C上有提到Kolmogorov–Smirnov test
主要用来衡量两个distribution的相似性
可以考虑使用K-S TEST | c**b
发帖数: 661 | 4 Lift chart, quintile analysis
【在 j***y 的大作中提到】
: 怎么比较2个不同的models哪个好关于policy loss，他们从同一数据来，不同的
: predictor. 不知道问题是不是这样，不是听的很清楚。
| A*******s
发帖数: 3942 | 5 my 2 cents, correct me if i were wrong:
This may be a general statistical question about variable selection. you may
need to compare r square, AIC, Cp of two models.
【在 j***y 的大作中提到】
: 怎么比较2个不同的models哪个好关于policy loss，他们从同一数据来，不同的
: predictor. 不知道问题是不是这样，不是听的很清楚。
| N****e
发帖数: 146 | 6 This could be a very complicated case.
In general, First,you need to clarify the goal of the project. eg,how
important the model result is, what is the financial impact of the model
result
Then, you need to figure out each model's cost (time frame,labor, data
requested, outsource requested, computer system etc) and the model accuracy.
last, decide the best option by obtaining the optimal balance from cost and
gain. | c****t
发帖数: 19049 | 7 it depends
if they are predictive models, it depends on which one minimize uinversal
errors;
if they are inference models, it can be told by comparing a lot of
statistical tests measures;
if they are just exploration/description models, it depends on how well they
catch the subtle changes in the dataset |
|
