최근 연구한 결과 가지고 실제 한국 주식투자에 적용할 수 있는 방법을 최대한 간략히 정리했다. 과거 성과는 압도적이다.
사게 되는 주식들을 보니까 진짜 사기 싫게 생겼다. 나는 한국에서 이런 주식들은 내려갈 땐 많이 내려가고 올라갈 땐 잘 안올라간다는, 경험과 여론을 통해 생긴 인식이 있었는데, 내 연구결과를 믿어봐야지... 주식을 사람들이 싫어해서 가격이 쌀 때 사는 가치투자가 원래 이런 거기도 하고.
과거의 성과가 미래의 수익을 보장하지 않는다는 펀드 약관 주의사항처럼, 미래 수익은 앞으로 몇년간 직접 내 돈 걸고 확인해볼 생각이다.
(2024. 11. 12. 추가) 오늘부로 마감한다. 여러 이유가 있는데, 일단 모델이 약간 틀린 곳이 있어서 수정을 해야 했고, 왜인지 모르겠는데 데이터가 없어져서 초기에 했던 작업을 처음부터 다시 해야 하는 상황이고, 한국 주식을 사기 싫기도 하고, 개인적으로 돈 나갈 데가 여기저기 꽤 있어서. 약 네 달간 거의 딱 10% 손실이고, 같은 기간 코스피는 10% 넘게 빠졌으니 그저 그런 (굳이 따지면 약간 나쁘진 않은) 성과다. 앞으로 미국 주식으로 하고 싶은데 데이터를 어디서 구하느냐가 문제다.
=====================================
I introduce a value investing strategy in Korea, using instrumented time-varying beta of the five-factor model of Fama and French (2015) (FF5 henceforth).
Data and methodology
The sample includes stocks listed in Korean Stock Exchange (KSE, also commonly known as KOSPI). I exclude stocks whose name contains at least one of the following words: '스팩', '은행', '지주', '보험', '리츠', '금융', '화재', '생명', '뱅크', '증권', ‘홀딩스’ since the intepretation of their accounting information should be different from stocks in other categories.
The procedure starts with calculating the daily and monthly FF5 return for the whole sample period. Next, for every month and stock, I calculate the daily forward FF5 beta of 100 working days. Now using the cross-sectional rank data of firm characteristics (e.g. accounting information and past-returns data), I obtain a mapping from the observable firm characteristics to the realized future beta of FF5 for each stocks, following Kelly, Moskowitz, Pruitt (2021).
At the beginning of each month, I estimate the forward FF5 beta of each stock by this mapping, and also estimate the expected return of FF5 factors with the recent 240 months average. Finally, I calculate the expected return of each stocks. Specifically,
where f denotes the mapping between the vector of cross-sectional rank of firm characteristics (C_(i,t)) and the realized future FF5 beta, and X ̅_t denotes the vector of recent 240 months average FF5 returns.
Trading strategy
The trading strategy is very simple: to buy the top 50 stocks with the highest expected return in equal weight at the beginning of each month. The reason I call this strategy as ‘value’ investing is because the returns of SMB and CMA is near zero in KSE, (the result would be different if I include the whole KSE+KOSDAQ market.) so this strategy tend to buy low-PBR stocks with high profitability. What distinguishes this strategy from common value investing strategies is that this strategy incorporates not only the accounting information of stocks, but it also includes the past return, price volatility, trading volume, etc. This works as the key of distinguishing value stocks with good momentum and market attention from other value stocks.
Past performance
During the 32-year period from 1992 to 2023, KOSPI increased by 4.5 times (CAGR 4.8%, denoted by the black line) while this strategy increased by 295.6 times (CAGR 19.5%, denoted by the red line). The monthly average return of the 10th decile portfolio is 2.03%, compared to -0.55% of the 1st decile portfolio.
To check more recent performance, during the 4-year period from 2020 to 2023, KOSPI earned 24% while this strategy earned 80%.
Momentum
Among the 50 stocks, I buy the 25 stocks with the highest return of month t-12 to t-2, the standard momentum signal. It appears from the past data that this filter improves the performance. (Korea is known to have no momentum effect... but maybe I found the cross-sectional stock-level momentum effect among value stocks..?)
During the 32-year period from 1992 to 2023, the strategy with momentum filter increased by 2,220 times (CAGR 27.2%, denoted by the green line).
Market timing
The difference of the leading economic indicator (LEI) of the two most recent months provided by the OECD is a good indicator of market downturn and volatility (source: https://www.youtube.com/watch?v=Csi_jJa3Vm4). Therefore, if the difference is below -0.2, I do not hold this strategy to avoid market downturn.
During the 32-year period from 1992 to 2023, the strategy with momentum plus market timing earned 4,146 times (CAGR 29.7%, denoted by the blue line). What matters more than the improvement of the return is the significant decrease in the volatility of the portfolio return (annual Sharpe ratio from 0.786 to 0.894).
Investment
From July 2024, I apply this strategy in actual investment with 20 million KRW. The rest of this document will be filled with the performance of this strategy from now on. I also update and share the portfolio of each month below.
202407 Portfolio
DL이앤씨 LX인터내셔널 TP 경동도시가스 넥센 다우기술 대성산업 대우건설 대한해운 두산 무림페이퍼 삼양사 상신브레이크 서연이화 성신양회 세방 신세계 지역난방공사 티에이치엔 한국가스공사 한국무브넥스 한진 현대제철 현대코퍼레이션 화승코퍼레이션
202408 Portfolio
DL이앤씨 DRB동일 DSR제강 GS글로벌 HDC HDC현대EP HL D&I SK디스커버리 TP 넥센 다우기술 대동 대한해운 두산 무림페이퍼 상신브레이크 세방 인터지스 지역난방공사 티에이치엔 한국가스공사 한국무브넥스 한진 현대글로비스 화승코퍼레이션
202409 Portfolio
DL이앤씨 DSR제강 HL D&I KCTC KSS해운 LS LX인터내셔널 계룡건설 다스코 다우기술 대한해운 두산 두올 상신브레이크 서연이화 세아제강 인터지스 자이에스앤디 지역난방공사 케이씨 티에이치엔 한국가스공사 한성기업 현대코퍼레이션 화승코퍼레이션
202410 Portfolio
DL이앤씨 DSR제강 HDC현대EP HL D&I KCTC LS LX인터내셔널 계룡건설 다우기술 대성산업 대우건설 대한해운 두산 무림페이퍼 삼천리 상신브레이크 인터지스 지역난방공사 티에이치엔 한국가스공사 한성기업 한솔로지스틱스 한진 현대코퍼레이션 화승코퍼레이션
References
Fama and French (2015) “A five-factor asset pricing model”, Journal of Financial Economics
Kelly, Moskowitz, Pruitt (2021), "Understanding momentum and reversal", Journal of Financial Economics