그런데 제가 구독하는 日経에서 보낸 메일속에서 재미있는 제목을 보았습니다. Line이 IBM의 Watson기술처럼 자체로 개발한 AI기술을 기업을 대상으로 판매한다는 내용이었습니다.
Line이 개발한 AI를 Line Brain이라고 합니다. Chatbot, OCR 및 Speech To Text가 제품화하여 서비스를 제공하는데 自然言語解析, 文字認識技術, 音声解析 및 画像解析과 관련한 AI기술을 서비스화한 것입니다.
Line은 각 서비스를 어떤 비전으로 접근하는지 살펴보겠습니다. 먼저 Chatbot입니다.
국내 인터넷은행들이 사용하는 주민등록증이나 운전면허증을 이용한 간편인증처럼 OCR기술은 eKYC에 활용한다고 합니다. 정형 문서뿐 아니라 비정형 문서를 이용한 OCR기술로 다양한 상거래에 적용할 계획이라고 합니다.
OCR技術の想定用途の1つが、オンラインで本人確認する仕組み「eKYC（Know Your Customer）」。利用者はスマホのカメラで運転免許証と自分の写真を撮影するなど比較的簡素な作業で、本人確認手続きを済ませられる。金融やEC（電子商取引）、C to C（消費者間取引）といったサービスを提供するうえで重要な技術だ。 LINEは自らもスマホ決済サービス「LINE Pay」にeKYCを導入済み。OCR技術を外販することで、同様な需要の高まりを取り込む。他にも家計簿アプリ向けのレシート撮影機能や、企業における領収書や請求書の電子化といった用途を見込む。
영상분석기술도 역시 상거래와 결합시킬 예정이라고 합니다. "音声認識や動画解析をフル活用した「次世代型テレビ」の構想"라는 제목처럼 음성인식기술, 동영상분석기술을 결합한 차세대 TV를 구상중이라고 합니다. 아울러 AI기술을 Line에 제공하는 서비스에 기반기술로 활용하는 전략도 취하고 있습니다.
솔류션으로 판매하는 Line Brain을 이용하여 자산관리시장에 도전해볼 수 있을까요?
이상과 같은 AI기술을 자본시장에 도입한다고 생각할 때 가장 먼저 떠오르는 것이 로보어드바이저입니다. 이제는 새롭지 않은 분야인 로보어드바이저가 생각보다 영향력을 키워나가지 못한 느낌입니다. 로보어드바이저가 각광을 받을 때 관심은 로봇이었습니다. 알파고의 영향때문인지 몰라도 로봇이 편드를 관리하면 더 높은 수익을 낼 수 있을 것이라는 막연한 믿음이었습니다. 이 믿음이 사라지는데 필요한 시간은 짧았습니다. 투자자들는 수익율을 기대했지만 기대보다 낮은 수익때문입니다. 핵심은 수익률이고 인공지능이나 기계학습을 기반한 투자가 각광을 받으려면 결국 시장에서 뛰어난 수익율을 보여야 합니다. 한국보다 앞선 해외의 사례는 어떤지 살펴보겠습니다.
2018년 봄에 쓰여진 Trust the machines? Funds run by artificial intelligence을 보면 AI를 기반으로 ETF가 둘 등장합니다. 첫째는 BIKR (Rogers AI Global Macro ETF)이고 둘째는 AIEQ( AI Powered Equity ETF)입니다.
ETF Managers Group and Ocean Capital Advisors launched an AI-powered fund last month dubbed the Rogers AI Global Macro ETF (BIKR) that invests primarily in single-country ETFs. The fund’s AI sifts through millions of data points from countries around the globe and uses what it learns to determine how best to allocate the fund’s holdings. (Humans carry out the trades, however.) Does this approach work? Another AI fund launched last November by ETF Managers Group and Equbot shows some promise.The fund, which is called the AI Powered Equity ETF (AEIQ), invests in a variety of U.S.-based companies and seeks to beat the returns of the S&P 500. So far, it’s getting it done. The ETF is up 8.1 percent this year, while the S&P 500 has gained about 1.5 percent.
"Humans carry out the trades, however"라는 문장으로 놓고 볼 때 두 ETF를 비교해보면 BIKR은 전담 포트폴리오 매니저가 있지만 AIEQ는 없는 듯 합니다. AIEQ나 BIKR가 어떤 알고리즘에 기반한 전략을 운용하는지 확인할 자료가 거의 없습니다. 그나마 참고할 만한 것이 SEC에 보고한 공시자료입니다. 먼저 BIKR입니다.
Principal Investment StrategiesThe Fund uses a “passive” or indexing approach to try to achieve its investment objective. Unlike many investment companies, the Fund does not try to “beat” the Index and does not seek temporary defensive positions when markets decline or appear overvalued other than those indicated in the Index. The Fund will use a replication strategy. A replication strategy is an indexing strategy that involves investing in the securities of the Index in approximately the same proportions as in the Index. However, the Fund may utilize a representative sampling strategy with respect to the Index when a replication strategy might be detrimental to shareholders, such as when there are practical difficulties or substantial costs involved in compiling a portfolio of equity securities to follow the Index, in instances in which a security in the Index becomes temporarily illiquid, unavailable or less liquid, or as a result of legal restrictions or limitations (such as tax diversification requirements) that apply to the Fund but not the Index.
The Rogers AI Global Macro Index
The Index was developed in 2018 by Ocean Capital Advisors®, LLC, the Fund’s index provider (“Ocean”). The Index tracks the performance of single-country (including emerging markets) exchange-traded funds (“ETFs”) that each track a broad-based index composed of equity securities primarily listed on an exchange in the applicable country (collectively, the “Underlying ETFs”) or an ETF tracking the 1-3 year U.S. Treasury Bond market (a “Treasury ETF”). The underlying index of each single-country Underlying ETF included in the Index may include equity securities of small-, mid-, and large-capitalization companies, and such equity securities are expected to have exposure to a wide range of industries reflective of the economy of the applicable country. The Index includes a single Underlying ETF per country. Each eligible Underlying ETF is specified
in the Index’s rules, and the Underlying ETFs generally reflect the ETF with the broadest exposure to equity securities in the applicable country and meeting certain minimum investibility criteria.The allocation of the Index’s weight among the Underlying ETFs or Treasury ETF is based on a proprietary artificial intelligence-driven algorithm (the “Model”) that analyzes macroeconomic data (e.g., volatility, interest rates, productivity, gross national product) monthly to identify likely changes in market directions in individual countries and within the global economy. The Model uses objective data to calculate the magnitude and probability of such market movements over an approximately 18-month period, while seeking to identify any “micro-cycles” that might develop in shorter time periods, to determine the optimal investment allocations based on the relative expected performance of the market in each country in the Index universe. The Model’s bias toward longer-term market movements is generally expected to minimize turnover at the time of each rebalance. When the Model determines to reduce or eliminate exposure to a country, the reduction to such allocation (whether all or a portion of the country’s allocation in the Index) is replaced with the Treasury ETF. From time to time the Index may be significantly allocated to the Treasury ETF. The Index is rebalanced as of the first business day of each month using data as of the last business day of the previous month. The weight of each country is capped at 10% at the time of each rebalance of the Index, with any excess weight reallocated equally to each other country included in the Index. At the time of each rebalance, the Index will have a positive allocation to at least twenty individual countries. The Index was developed by Ocean for purposes of creating the Fund. The Index is calculated and maintained by Solactive AG, an independent third-party calculation agent. As of January 14, 2019, the Index consisted of single-country ETFs representing 43 countries. The four largest country allocations in the Index were Brazil (6.89%), China (4.59%), Malaysia (3.65%), and Nigeria (2.79%), and 48.37% was allocated to the Treasury ETF.
다음은 AIEQ입니다 AIEQ는 Equbot이 개발하였고 IBM의 Whatson기술을 채택하여 만든 ETF입니다.
Principal Investment StrategiesThe Fund is actively managed and seeks to achieve its investment objective by investing primarily in equity securities (or depositary receipts) of companies in developed markets outside the United States based on the results of a proprietary, quantitative model (the “EquBot Model”) developed by EquBot LLC (“EquBot” or the “Adviser”) that runs on the Watson™ platform. EquBot, the Fund’s investment adviser, is a technology based company focused on applying artificial intelligence (“AI”) based solutions to investment analyses. As an IBM Global Entrepreneur company, EquBot leverages IBM’s Watson AI to conduct an objective, fundamental analysis of companies in non-U.S. developed markets based on up to ten years of historical data and apply that analysis to recent economic and news data.
Each day, the EquBot Model ranks each company based on the probability of the company benefiting from current economic conditions, trends, and world events and identifies approximately 80 to 250 companies with the greatest potential over the next twelve months for appreciation and their corresponding weights, while maintaining volatility (i.e., the range in which the portfolio’s returns vary) comparable to the broader non-U.S. developed market. The Fund may invest in the securities of companies of any market capitalization.The EquBot Model recommends a weight for each company based on its potential for appreciation and correlation to the other companies in the Fund’s portfolio. The EquBot Model limits the weight of any individual company to 10%. IBM’s Watson AI is a computing platform capable of answering natural language questions by connecting large amounts of data, both structured (e.g., spreadsheets) and unstructured (e.g., news articles), and learning from each analysis it conducts (e.g., by recognizing patterns) to produce a more accurate answer with each subsequent question. The Adviser utilizes the recommendations of the EquBot Model to decide which securities to purchase and sell, while complying with the Investment Company Act of 1940 (the “1940 Act”) and its rules and regulations. The Adviser anticipates primarily making purchase
and sale decisions based on information from the EquBot Model. The Fund may frequently and actively purchase and sell securities.
이외에 한국 Qraft와 EXCHANGE LISTED Funds가 제휴하여 QRFT를 상장하였습니다. SEC에 보고한 문서가 담고 있는 전략입니다.
Principal Investment StrategiesThe Fund is an actively-managed exchange-traded fund (“ETF”) that seeks to achieve its investment objective by investing at least 80% of its net assets, plus the amounts of any borrowings for investment purposes, in securities of U.S.-listed large capitalization companies. The Fund defines large capitalization companies as companies having a market capitalization in excess of $4 billion at the time of purchase. The Fund will invest in equity securities of such companies, including common stock, American Depositary Receipts (“ADRs”) and Global Depositary Receipts (“GDRs”). The Fund’s adviser, Exchange Traded Concepts, LLC (the “Adviser”) uses an investment process based on a proprietary artificial intelligence security selection process that extracts patterns from analyzing data, as discussed below, developed by QRAFT Technologies, Inc. (“Qraft”). Qraft is a South Korea-based provider of artificial intelligence investment systems and currently offers services to various financial institutions in Korea. The Adviser has licensed Qraft’s proprietary artificial intelligence security selection process for the management of the Fund.
In pursuing the Fund’s investment objective, the Adviser consults a database generated by Qraft’s AI Quantitative Investment System (“AQUA”), which automatically selects and weights portfolios of companies to provide a balanced exposure to five main factors affecting the U.S. market (the “Five Factors”): quality (generally, a company’s profitability), size (market capitalization), value (comparison of a company’s market value versus its book value), momentum (a security’s recent price returns compared to the overall market over time), and volatility (a security’s systematic risk as compared to the market as a whole) (the “U.S. Large Cap Database”). In creating the U.S. Large Cap Database, AQUA utilizes automated data feed and data processing using deep learning technologies (i.e., exposure to and processing of large amounts of data). First, on a monthly basis, AQUA automatically sends queries to and collects from various data vendors company fundamental data (such as historical stock prices and other financial information) of all companies listed on the New York Stock Exchange and NASDAQ as well as macroeconomic data. AQUA then processes and stores newly received data with stored historical data. AQUA processes such data first by creating five indices each representing one of the Five Factors of companies in the top 20% of U.S.-listed companies based on market capitalization. Then, AQUA compresses such data over the last 60 month period and, using deep learning technologies, evaluates how each individual factor would change and/or affect a company over time and identifies 300 to 350 companies that have the greatest potential to outperform their U.S. large cap peers over the next three-month period. AQUA estimates a weighting for each such company based on its potential for maximum return as compared to other companies, and the final portfolios are then delivered to the U.S. Large Cap Database for use by the Adviser. AQUA repeats such processes on the first business day of every month and the Adviser makes or changes investments in the Fund based on the newly generated information.
The Fund expects to hold 300 to 350 companies in its portfolio. While it is anticipated that the Adviser will purchase and sell securities based on recommendations by the U.S. Large Cap Database, the Adviser has full discretion over investment decisions for the Fund. Therefore, the Adviser has full decision-making power not only if it identifies a potential technical issue or error with the U.S. Large Cap Database, but also if it believes that the recommended portfolio does not further the Fund’s investment objective or fails to take into account company events such as corporate actions, mergers and spin-offs. Additionally, the Adviser has discretion over the amount of cash maintained in the Fund’s portfolio and the reinvestment of dividends in the Fund’s portfolio, subject to the Fund’s distribution requirements as a regulated investment company for federal income tax purposes. See “Federal Income Taxes” in the Fund’s Statement of Additional Information (“SAI”) for a more complete discussion. Notwithstanding the foregoing, the Fund limits the weighting of a single company to 10% and no more than 40% of the Fund’s assets may be invested in securities with a more than 5% weighting in the Fund’s portfolio. Because the U.S. Large Cap Database is adjusted on a monthly basis, the Adviser expects that the Fund will frequently purchase and sell shares of securities.
기본적으로 Active와 Passive의 차이가 존재합니다.
액티브 펀드(Active Fund) : 시장 수익률을 초과하는 수익을 올리기 위해서 펀드매니저가 적극적으로 시장이나 종목의 가격을 예측하여 좋은 종목을 적절히 매수, 매도하는 펀드. 패시브 펀드보다 운용비용이 더 들고, 기대수익이 높은 만큼 변동성과 리스크도 상대적으로 높음. 패시브 펀드(Passive Fund) : 대표주가지수를 구성하는 종목들로 펀드를 구성해 그 지수가 오르는 만큼의 수익을 추구하는 펀드. 시장이 효율적이라고 믿기 때문에 펀드매니저의 주관적인 운용이나 개입을 필요로 하지 않으며 보수가 저렴함.
그외 Qraft는 팩터모델에 Deep Learning을 적용한 방식입니다. AI ETF의 제작 및 NYSE 상장, 한달동안의 성과이 자세히(?) 소개하고 있지만 위 인용문중 Bold로 처리한 부분입니다. 반면 AIEQ은 거의 모든 상장종목을 분석하여 종목을 선택하는 방식인 듯 합니다. EquBot Taps AI To Perform The Work Of 3,000 Research Analysts이나 SEC 인용문에서 Bold로 처리한 부분입니다.
The EquBot model, with the help of powerful computing platforms like IBM Watson, compiles and computes over a million data points daily, from quarterly reports, news articles, social media posts, financial statements and more. The model then builds predictive financial models for thousands of domestic and global companies to identify timing and positioning for optimal stock selection and potential capital appreciation. We've set the fund's objectives to meet commensurate volatility measures and have so far delivered benchmark-beating returns in both ETFs.
AIEQ와 QRFT의 차이점은 데이타처리주기입니다. QRFT는 월단위로 이루어지는 반면 AIEQ는 일단위로 이루어집니다. 그렇기 때문에 처리하는 데이타도 차이가 많이 납니다. Rebalancing도 월단위, 일단위로 이루어집니다.
사실 AI ETF에 대한 비판도 많습니다. 그 중 기계학습을 리서치에 도입한 New Constructs가 제시한 비판입니다. Robo Analyst라고 하며 HBR은 New Constructs: Disrupting Fundamental Analysis with Robo-Analysts라는 사례보고서도 내놓은 회사입니다. Don’t Believe the Hype About this “AI Powered” ETF중 일부입니다.
This section raises a number of questions, such as:What data points compose the “one million pieces of information” being analyzed daily? Where does AIEQ get the data from, and what processes are in place to ensure data quality? What is the methodology behind its “predictive models”? How does the AI process the data to decide which companies have “high opportunities of capital appreciation”?
그러면 New Constructs이 제시하는 방법론은 무엇일가요? Origin of the Robo Analyst에서 자세히 소개하고 있습니다.
[gview file="http://smallake.kr/wp-content/uploads/2019/08/newconstructsintro.pdf"]
인공지능과 관련한 기술이 있다고 하면 어디에 적용할까요? 개인적으로 자산관리는 추천하지 않습니다. (^^)
