1.
자주 나왔던 주제입니다. 이번 논문은 키워드의 흐름과 주가를 비교하였습니다.
미국 경제전문매체 비즈니스 인사이더(Business Insider)는 미국 보스턴 대학·영국 워릭 대학 물리학·행동과학 공동 연구진이 구글 검색으로 다음에 찾아올 금융위기를 알아낼 수 있다는 연구결과를 발표했다고 29일(현지시각) 보도했다.
연구진은 지난 2004년부터 2012년 사이 미국 내 주요 대기업 500개의 S&P500지수(Standard & Poor’s 500 Index) 변동흐름과 동 기간 내의 구글 검색 키워드 변동흐름을 비교하는 데이터 분석을 최근 진행했다.
결과는 흥미로웠다. 해당 시기 중 미국 증권시장이 하락세로 접어들기 직전 구글 검색 키워드의 대부분은 ‘산업’과 ‘정책’ 관련 주제로 꽉 차있었다. 음악, 날씨처럼 주가 변동과 별 관련이 없는 검색어는 거의 관찰되지 않았다.
‘구글링’으로 금융위기 미리 예측가능…어떻게?중에서
기사는 Your Google Searches Could Predict The Next Stock Market Crash을 번역하였습니다. 기사가 전한 논문은 아래입니다.
Quantifying the semantics of search behavior before stock market moves
논문이 제시한 결론도 그동안의 논문과 다르지 않지만 다른 분야에도 적용할 수 있다고 강조합니다.
The method we present here facilitates in a number of ways the interpretation of the relationship between search data and complex events such as financial market moves. First, the frequency of searches for a given keyword can grow and decline for various reasons, some of which may or may not be related to a real-world event of interest. This method allows us to abstract away from potentially noisy data for individual keywords and identify underlying semantic factors of importance. Second, our method allows us to extract subsets of search data of relevance to real-world events, without privileged access to full data on all search queries made by Google users. By identifying representative keywords for a range of semantic topics, such analyses can be carried out despite limitations on the number of keywords for
which search data can be retrieved via the Google Trends interface. Third, our semantic analysis is based on simple statistics on how often words occur in documents alongside other words. As a result, the analysis presented could be carried out in languages other than
English—for example, using other editions of Wikipedia—with no extra modifications to the approach required. We suggest that extensions of these analyses could offer insight into large-scale information flow before a range of real-world events.
2.
그러면 이전 연구중 어떤 것이 있을까요?
Quantifying Trading Behavior in Financial Markets Using Google Trends
Twitter Data Analytics에서 소개한 논문이기도 합니다. 이 논문을 기초로 하여 매매알고리즘을 개발한 분이 계시네요.
Google Search Terms predict market movements
Quantopian에 올라온 소스입니다. Python입니다.
|
1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 27 28 29 30 31 32 33 34 35 36 37 38 39 40 41 42 43 44 45 46 47 48 49 50 51 52 53 54 55 56 57 |
# This algorithm recreates the algorithm presented in # "Quantifying Trading Behavior in Financial Markets Using Google Trends" # Preis, Moat & Stanley (2013), Scientific Reports # (c) 2013 Thomas Wiecki, Quantopian Inc. import numpy as np import datetime # Average over 5 weeks, free parameter. delta_t = 5 def initialize(context): # This is the search query we are using, this is tied to the csv file. context.query = 'debt' # User fetcher to get data. I uploaded this csv file manually, feel free to use. # Note that this data is already weekly averages. fetch_csv('https://gist.githubusercontent.com/twiecki/5629198/raw/6247da04bacebcd6334a4b91ed21f14483c6d4d0/debt_google_trend', date_format='%Y-%m-%d', symbol='debt', ) context.order_size = 10000 context.sec_id = 8554 context.security = sid(8554) # S&P5000 def handle_data(context, data): c = context if c.query not in data[c.query]: return # Extract weekly average of search query. indicator = data[c.query][c.query] # Buy and hold strategy that enters on the first day of the week # and exits after one week. if data[c.security].dt.weekday() == 0: # Monday # Compute average over weeks in range [t-delta_t-1, t[ mean_indicator = mean_past_queries(data, c.query) if mean_indicator is None: return # Exit positions amount = c.portfolio['positions'][c.sec_id].amount order(c.security, -amount) # Long or short depending on whether debt search frequency # went down or up, respectively. if indicator > mean_indicator: order(c.security, -c.order_size) else: order(c.security, c.order_size) # If we want the average over 5 weeks, we'll have to use a 6 # week window as the newest element will be the current event. @batch_transform(window_length=delta_t+1, refresh_period=0) def mean_past_queries(data, query): # Compute mean over all events except most current one. return data[query][query][:-1].mean() |
