p********0 发帖数: 186 | 1 Hi,
Does anyone have experience with the Global Equity Model/MSCI-GEM.
It seems that the underlying assumption/intuition is not valid.
Assume we have 2 countries in the universe for simplicity reason. US and
China.
Then we have 2 risk factors
1) Country
2) Value/Growth.
Assume US Growth company outperform Value, but Chinese Growth company
underperform its peers.
If we run regression for US only companies, we will see a positive return on
the Value risk factor part. However if we run regression for US and Chinese
companies as a whole, the return for the Value Factor will be close to zero
, offset by the Chinese companies.
Since we know stocks perform different in the local market, does running
regression on the whole global market make sense? | u****l 发帖数: 5008 | 2 I don't know the exact context, but seems you can use a difference-in-
difference regression specification.
http://www.nber.org/WNE/lect_10_diffindiffs.pdf | w**********y 发帖数: 1691 | 3 没看过GEM,但是懂一点factor model。
难道GEM里不是每个single security对应每个factor都有自己不同的factor loading (
beta, exposure, whatever you call it)?
比如你每个公司 i 的value factor: F 定义的是book-to-price ratio,应该每个公
司都有自己的 bi吧?
【在 p********0 的大作中提到】 : Hi, : Does anyone have experience with the Global Equity Model/MSCI-GEM. : It seems that the underlying assumption/intuition is not valid. : Assume we have 2 countries in the universe for simplicity reason. US and : China. : Then we have 2 risk factors : 1) Country : 2) Value/Growth. : Assume US Growth company outperform Value, but Chinese Growth company : underperform its peers.
| p********0 发帖数: 186 | 4 Each company has a factor loading
R_i = factor_i_j * factor_premium_j, i is the index for company, j is the
index for factor assume total M factors.(R_i is the return for last one
month)
the factor_premium_j should be the same for all companies.
Known R_i, factor_i_j(factor loading) for i=1, 2, ,, N and J=1,2,,,,N
factor_premium_j is calculated using regression, but this is based on
premise that all global companies have same factor_premium_j, which is not
true. US Value premium is not the same as Japan/China Value premium for the
last month. | w**********y 发帖数: 1691 | 5 好像barra model是分 US, Euro和global的吧?
我做过类似的global factor model, 每个industry和region都处理成一个factor (
premium as you said), loading matrix is just a sparse matrix.
the
【在 p********0 的大作中提到】 : Each company has a factor loading : R_i = factor_i_j * factor_premium_j, i is the index for company, j is the : index for factor assume total M factors.(R_i is the return for last one : month) : the factor_premium_j should be the same for all companies. : Known R_i, factor_i_j(factor loading) for i=1, 2, ,, N and J=1,2,,,,N : factor_premium_j is calculated using regression, but this is based on : premise that all global companies have same factor_premium_j, which is not : true. US Value premium is not the same as Japan/China Value premium for the : last month.
| l*****n 发帖数: 1333 | 6 There has been some emperical research showing that certain factors work
across countries, e.g. Value, http://schwert.ssb.rochester.edu/f532/JFE12_FF.pdf
..and therefore the rationale to estimate style factors globally. But I
agree with you, especially over shorter horizons, factors behave differently
in different countries. | C******n 发帖数: 9204 | 7 这个literature好像叫global fama-french。。。
中国有个额外factor,和流通股数量有关。而且中国size effect大,value effect不
明显。 | o********n 发帖数: 100 | 8 好奇问一下fama-french管用吗?
我们都用hidden factor model至少100个factor的。。。 |
|