"Vitruvius Curve" 어록 (on work)

Vitruvius Curve가 어떤 사람인지는 모르겠지만 트레이딩에 대해 깊이있는 글을 썼다. 아이디가 참조하고 있는 비트루비우스는 보통 다빈치의 비트루비우스적 인간으로 잘 알려져있다. 예전 미학 수업에서 건축의 원칙으로 비례를 쓴 게 비트리비우스이고, 거기서 영감을 얻어 다빈치가 그 유명한 그림을 그렸다고 들은 기억이 난다. 정확하진 않다.

다수의 글이 잘 이해가 되지 않지만, 어떤 글들은 깨닫게 해주는 바가 많다. 두고두고 배울 점이 많지 싶다.

Vitruvius_Book tweets pdf.pdf

1.01MB

 

---

발췌
-AI/ML전략에 뒤쳐질까하는 걱정은 아직은 이를 수 있다. 실력이 매우 뛰어난 AI/ML 전문가에게 3년동안 재량을 주었지만 문제점들이 있었다. 1.미래 수익 분포는 무조건 바뀐다는 것을 수용할 수 없는 것 2.도메인 지식과 경험에서 연구하기보다는 알고리즘으로 전략을 찾으려 하는 것 3. 전략이 모멘텀과 밸류에이션 측면을 동시에 가져가야하는 것 4. 자기들이 낸 연구에 의존하는 것 5.안정적인 분포를 찾는 것에는 매우 뛰어나지만..

I am writing a fair bit on AI/ML with awareness that this is of value to non AI/ML players.
There is quantitative and there is quantitative. My firm is fully quant/process driven. A separate team that I hired is pure artificial intelligence and machine learning they were all "let go". I backed 5 world class practitioners and gave them free rein, three years and anything they needed. We tried to analyse why they just couldn’t predict. 1. They couldn’t accept or prove to their satisfaction that the future distribution of returns is guaranteed to change. 2. They wanted the algorithms to find their strategies rather than applying experience and domain knowledge with first principles research. 3. They couldn’t accept that a strategy must accept BOTH momentum and reversionary outcomes of future price movement. 4. They were too reliant on their own precious published work rather than alpha research. 5. At its core - some of their work was absolutely outstanding for phenomena that exhibit PERMANENTLY stable distributions, but for scenarios where it is possible/likely that the fourth moment may/does diverge - no good. All went to an engineering firm.

-모멘텀과 밸류에이션  (*주: 원문 표현은 reversionary, 평균회귀의 뜻에 가깝다. ) 모두에서 이득을 취할 수 있도록 해야 한다. 모멘텀 장세에서는 모멘텀을, 밸류에이션 장세에서는 밸류에이션을 따르라. 이는 쉽지 않지만, 있는 그대로의 시장을 봐야한다.

In my experience, a trader’s risk taking matrix needs to include  strategies that can extract profits when the market is rewarding reversionary outcomes and make money when the market is paying momentum traders. This is not easy to do technically and goes against much market folklore, bias and ’hero worship’. Look at it this way - if the market under consideration, for the holding period concerned historically, presently and in your forecast demonstrates reversionary behaviour, then why fight that? Why deflect with sentences like "I only trade with the trend"? Equally and similarly, if the market under consideration, for the holding period concerned historically, presently and in your forecast demonstrates ’momentum’ or trend behaviour[not that I can define either in advance but that is beside the point] then deploy momentum approaches. Why deflect with sentences like “...its too overbought / oversold” [again terms that I cannot define in 25 years of research]. When a superstar trader says that I don’t fight the markets, this is what they mean. Ignore should you wish, but this quantitative and cerebral flexibility is priceless in a PM.

-시장의 큰손들만이 아는 규칙이 있을까? 없다. 나는 25년간 수많은 거래를 했고 수익률을 상승시켰지만 아직 모멘텀도 정의하기가 어렵다. 모멘텀은 후행적으로만 정의된다.

Here’s one for the new players sitting there wondering what are the great secrets the big players know? We have done a LOT of work over the past two decades - on and off - on runs versus reversals in price and other data. Some has been very fruitful and has helped us make more money or helped us opportunistically reduce the volatility of our returns. I feel that all this work qualifies me to say one slightly contradictory statement: I cannot define what a trend is until it has already happened! For the avoidance of doubt, we are almost 100,000 trades live so we are not just sitting there wringing our hands about definitional issues. But this is a matter of consequence. The greatest model that will ever be discovered is the one that a/ defines a ‘trend’ before it occurs and b/ allows one to have a reasonable idea of its genesis and demise with that knowledge (trend or reversion) one can apply exceedingly robust risk taking techniques without the pitfalls extant in markets as we look at them today! Now you know. We do not know.

-알파를 만들기 위해, 초심자는 아주 단순한 것부터 시작해야 한다. 연속적인 거래의 run과 sequence를 보라... 중략

One kernel for Alpha hunters...  I would like to share with those wishing to hear, the basic experiments that I did that led to finding phenomena and regularities that I was able to systematise and apply in markets...It is important to 8 understand that there are no advanced concepts to be applied in the embryonic stage of alpha research and so "some will be frustrated at the simplicity". Experiment No. 1 We need to understand the difference between reversion and momentum. The easiest way to think about this is to look at runs and sequences of consecutive transactions in a market. It is instructive to study what is happening as run lengths increase and change sign. A less taxing and simpler still method to study reversion versus momentum is to study the results from a three period moving average over relatively high frequency data (say 5 min). Again you will see runs (trends) and reversals...(choppy price action where trading the average gets you killed). Note and observe what happens to price both as runs increase in length and accelerate and what happens as markets experience strong reversion. Watch it long enough and something very obvious will occur to you...

-모두가 다 참여하고 모두가 쓰는 지표는 보잘 것 없다. 접근성이 제한된 시장에서 특정 플레이어들의 움직임을 볼 수 있는 지표는 알파가 될 수 있다.

-백테스팅 좋아하는 사람들에게: 매일 이전에 없던 새로운 가격 움직임이 생겨나기에 복잡도가 곱절이 되고, 백테스트의 가치는 시간이 흐를 수록 급격히 하락한다.

Backtest-ers for Alpha Consider : From a pedagogical standpoint, the reason why the efficacy of any backtest rapidly declines hyper- proportionally to time is that each market EverySingleDay does something, outputs some combination of price change, that it has NeverDoneBefore. I operate under the view that errors in financial markets multiply (rather than being additive). This makes the computational problem of high impact prediction increasingly complex each time a market acts as it NeverHasBefore 

-알파를 찾아 트레이딩하는 것은 대체로 아주 오랫동안 좌절스러운 과정을 거치고, 그 다음 찾아내서 기쁨을 느끼고, 그 다음에는 루틴하게 돌리는 일이다. 
Regardless of style, technique or classification - and real time implementation is mostly about long periods of frustration, moments of exhilaration and then routine. It’s an interesting circular feedback loop. There is little else to match the exhilaration of discovering a regularity/proving it is real and implementing it with real money (in the world of trading anyway). I wonder though if the best ideas are not paid for with frustration. In my case it’s about 500:1 but I’d still take that angst to trade alpha others don’t have.
-딱 하나의 아이디어를 밀어붙이라는 조언
My mentor used to say that we should find the ONE idea that has the world in its grip and trade markets in the opposite direction. ‘High’ valuations, the end of the world pandemic and alleged Central Bank largesse are major ideas that have had the world in their respective...grips. One might even say that one should live life in the opposite direction to the idea that has the world in its grip! It’s a helpful philosophy to assist one ignore the latest ‘cause du jour’ This is not a comment on today’s markets. Only a note for your future benefit.

-투자 스타일이 획일화된 신입들만 온다. 매크로, 롱숏 주식, 추세 추종부터 이제는 퀀트까지 다 똑같은 내용만 배워서 온다. 엘리트 프로그래머가 비 프로그래머보다 더 잘할거라는 믿음은 당혹스럽고, 회사는 알려진 전략만 연구하도록 허락해준다. 

For my sins, I am ‘experienced’ enough to have seen all of ‘Macro’, ‘L/S Equity’ and ‘Trend Following’ implode inward to varying degrees (Clearly, firms have survived in all three).One notes, regrettably, that it is beginning in the ‘Quant’ space. I believe I am qualified to state with some measure of certainty, the reasons- * Most new entrants come from the same echo chamber schools with the same knowledge. 14 * Very, very few have actually bought or sold anything in their lives. * A completely baffling belief that being a world class programmer means they will find better systematic alpha than a ‘non programmer’ given excellent testing software. * Firms only allowing research into published strategies (WHAT THE F.)

-샘플 사이즈가 하나인 이벤트에 대해 과신하는 경향 

The ending of a sentence describing a move in a financial product/indicator/data with the word ‘....ever’ means - by definition - that the sample size for the event is one. A speculation in markets based upon it and it alone has no repeatable predictive efficacy.The ‘wings’ of observation in financial markets and related data haves tended to diverge from centrality based on exogenous events as time has gone on. Whether or not variance is infinite is another question altogether.

-변동성은 약자에게서 강자에게 돈을 옮기기 위해 존재한다. 미래에는 네가 관측한 것보다 많은 변동성이 있을 것을 예상해라. 특정 금융시장의 곱셈적 특징은 이미 알려진 금융시장의 분포 교리를 망가트린다 

Three thoughts that I think are very relevant for financial markets speculation: 1. Volatility exists to take money from the weak and transfer it to the strong. 15 2. Expect more variance in the future - from data that does not yet exist - than you have observed in testing. 3. The multiplicative nature of certain financial market permutations and combinations blows all published financial market distributional orthodoxy’s out of the water. An enterprising researcher/trader/portfolio manager might even quantify these concepts for personal gain.

-알파 찾기 : 경로의존성, 동일 방향으로 움직이는 종목들 선후행성 보기

Alpha Seekers - Understanding Path Dependency
 Here is an interesting (although heavily contrived) experiment that is very helpful in *actually understanding* path dependency and, done often enough, can lead to some very good ideas. Take three markets that, on a close to close basis, tend to have a correlated ‘sign’ (+ or -). Put another way, all three tend to close in the same direction over time. Maybe EURUSD, GBPUSD & AUDUSD. What’s intriguing from a research angle is the different paths...that they take between the opening and the close. (Or any two points in time really) Things worth trying to quantify: * Leads and lags between them. * Absolute and relative sizes of moves. * Reactions to ‘X’ time period highs and lows. * Is EURUSD more likely to lead because it is ‘more important’ than the AUDUSD. * If all three move in lockstep with one another is that predictive of anything. * If one currency plays catchup to the other two late in the session, is that predictive of anything. And on...Not suggesting any of the above have any relationship to anything, rather these are a few things that come to one’s mind straight away as measureable, testable and replicable. You might pick three stocks in the same sector also. Ed. Note- one more dimension lower into the microstructure and it is quite evident to see cross hedging activities by liquidity providers that provide much nourishment for the HFT community (amongst many other morsels)

-퀀트 트레이딩에 대한 두려움. 퀀트 트레이딩을 이길 다음 전략을 못 만들 수도 있다는 두려움이 든다. 숙련된 재량적 투자자들이 퀀트 룰을 써서 높은 수익률을 기록할 수 있지 않을까. 

One of the things keeping me up at night is my impending  dread regarding ‘Quantitative Trading’. I have not noticed any particular diminution in what we do (other than high frequency) but I know for a fact that my work doesn’t exist elsewhere. I need to think about the next big strategy and the possibility (indeed the likelihood) that I won’t discover it. I have posted about a massive ML/AI experiment that didn’t succeed (not giving up but. . . tik tok, tik tok). Despite my background and quant trading bona fides, I am increasingly drawn to the Idea that the highest future returns will come from quantitative trading rules applied by experienced discretionary traders. This might not be the sexy ‘silicon based life form’ outcome that is being talked about but, increasingly, I believe a kind of ’reverse takeover’ may occur in which ‘quant’ will be - If not taken over - then ‘absorbed’ into the discretionary framework. I am still in the early stages of thinking about this but wouldn’t it be interesting if all the advances made in Quant - in the final analysis - merely served to make human traders better rather than destroy them.

-*1.어떤 시장이든 테스트하기 위해서는 모멘텀부터 정의하곤 한다.2.시장은 낮은 확률 시나리오 베팅에 수익을 준다. 높은 확률 시나리오 베팅에 수익을 주기전까지. 그 때 저위험 회귀가 나타난다.. 5.가장 좋은 진입점은 가장 불편한 시간들이다. 이건 피할 수 없다.,7.자 시계가 짧아질 수록 포트폴리오 관련 지표들( 분산 투자, 공분산, 시장 중립, 변동성/위험 조정 등) 은 성과에 오히려 독이 된다.  

Alpha Hunters...  conundrums: 1. I need a definition of the word ‘Trend’ for any market to test. 2. Prices move but if your risk tolerance is the same as the market then normal path dependency or an exogenous event will take you out. 3. The market rewards those trading low probability scenarios (long smooth trends) until it decides to reward higher probability, lower risk reward reversionary approaches. The trick is in the switch. 4. PM’s/Traders/‘Investors’ MUST take risk. There is less career risk implementing your strategy at full size than there is hiding under your mothers skirts. 5. Often the best entries occur at the most inconvenient times. This must be embraced. 6. Often discontinuous moves/fat finger levels/moves on extreme news/flash crash lows etc... tend to be hot days or weeks later - almost as IF that is where the market wanted to go anyway. 7. For shorter term time frames (and increasingly longer term holding periods) words like Diversification, Co-Variance, Market Neutral, Vol/Risk adjusted are - in fact - the enemy of performance.

-투자 스타일 무관, 일련의 금융 이벤트에 대한 공부를 하지 않은 사람과 대화하긴 어려웠다. (*주: 닉슨 금태환 폐지 이전 사건은 전혀 모르겠다) 모든 것에 관한 정확한 통계적, 처음의 기록이 남아있지만 다른 것들은 학위가 필요할 수 있다. 

I find it difficult to converse with quantitative, discretionary orgut feel traders/analysts who have not studied: The great stock and grain attempted (and successful) corners of the late 1800’s. The real and alleged activities of one JP Morgan in the 1920/30’s...The ‘real’ tech/stock mania and crash between, say 1968 and 1974. The move in the USD in the 1970’s after President Nixon severed the $35 nexus to Gold... (Also the reactions of the European financial sector and the eventual actions by the Swiss). FED Chairman Volker’s actions to kill inflation and lower the long term cost of capital for the US government for the next 38 years. The early 1970’s commodity moves (not just oil). The great currency ‘accords’ of the 1980’s. Secretary Rubins decision to commence a strong dollar policy, inflate the yield curve and save the world after the US savings and loan crisis 89/92 (Later repeated on a global scale with 2007/08 debacle). There exist reasonably accurate statistical and first hand records of everything since then but the other stuff - well, Wikipedia just won’t do- it will take some scholarship. What use to a quant? If you are asking that question then I think you have already lost the game. (Also the early 1990’s ERM and late 90’s Asia crisis).

-이하 규칙에서 임의의 값을 쓰고 있다면 스스로를 되돌아보자. 1. 시가 거래 2. 종가 거래. 3.VWAP같은 전략을 실행에 쓰는 것 4.수익/손실 시 포지션 '절반'을 청산하는 것 5.공짜 데이터나 실행?을 쓰는 것 6. 미래 가격 분포에 대한 감이 없거나 NLP, latency 역량 없이 경제 지표, 데이터를 매매에 쓰는 것 7.오늘과 어제 같은 방식으로 거래하는 것 

If you use any of the following arbitrary inputs in your process- ask yourself why? 1. Dealing at the Open. 2.Dealing at the Close. 3. Using low footprint strategies like VWAP for execution. Strategies that are in fact the antithesis of adding alpha during the execution phase. 4. Cutting 1/2 of your position when in profit or loss (why half? Why not 16%). 5. Relying on free data or execution. 6. Trading economic fundamentals/data release strategies without NLP and latency advantages and without proving to yourself that the data has anything whatsoever to do with the future distribution of prices. 18 7. Treating today the same as yesterday in terms of, for example, stop loss magnitude.