《findings alphas》Reading Notes

PART 1 Intriduction

1 Indroductuin to Alpha Design

• An aplha is a combination of mathmatical expressions, conputer source code, and configuration parameters that can be used, in combination with historical data, to make predictions about future mobements of various financial instruments
• the ratio of return to standard deviation of the returns is the information ratio of the alpha

2 Alpha Genesis-The Life-Cycle of a Quantitative Model of Financial Price Prediction

• The markets are an aggregate of people’s intensions, affected by changing technology, macro-economic reality, regulatios and wealth - which makes the bussiness of prediction more challenging than meet the eye

• DATA -> APLHA -> PRICE PREDICTIO

• some pointers to alpha evaluation

  • good in-sample performace doesn’t guarantee good out-of-sample performance
  • just like in academic statistics, outliers can ruin a model and lead to erroneous predictions.

• LOOK BACK

  • history never repeats itself exactly, ever. So while an alpha idea may look great on paper, there is no guarantee it will continue to work in the future
  • it’s easy to look back at history and imagine that the market was easier to trade than it was in reality

• the nature of the alphas changes with time, it is researcher’s challenge to find which ideas are relevent today, and to implement them in a way that is efficient, robust, and elegant

3. cutting losses

• no rule ever works perfectly. We call it the “UnRule”
• Applying all rules simultaneously is the key to success
• following implications:
  • It is necessary to come up with as many good rules as possible
  • No single rule can ever be relied upon completely
  • It is necessary to come up with a strategy for using rules simultaneously

Part2 Design and Evaluation

4 Alpha Design

• categorization of alphas
  • Intraday alphas
  • Daily alphas
  • weekly/monthly alphas
• Development of an Alpha(typical source information）
  • Price/Volume
  • Fundamentals
  • Macro data
  • Text
  • Multimedia
• Practical Alpha Evaluations
  • Information ratio: The mean of alpha’s returns /the standard deviation of the returns
  • Margin: The amount of profit made by the alpha divided by the amount of trading
  • Uniqueness: maximun correlation of the alpha to others in the (Lower correlation tend to mean that the alpha is more valueable)

5. How to Develop an Alpha. I: Logic with an Example

• The goal is to make profits while minimizing risk
• steps:
  • steps1: collect information
  • steps2: come up with an idea
  • steps3: translate into a mathematical expression
  • step4: transform the raw expression by applying operations
  • step5: final robust alpha
  • step6: translate into positions in a financial instruemnt
  • step7: check for robustness

6. How to Develop an Alpha. II: A Case Study

• The most widely used definition is: a computer algorithm used to predict financial instruments’ future movements
• alpha1 = -(close(today)-close(5_days_ago)/close(5_days_ago)
• alpha2 = alpha1, Sum(Aplha2 value within same industry)=0
• alpha3 = rank(appha1), Sum(Aplha3 value within same industry)=0
• new alpha = new alpha + weighted_old_alpha

7. Fundamental Analysis

• Main source of financial statements are the balance sheet, income statement and the statement of cash flows
• In-depth analysis of financial statements gives us insight into a company’s current and future performance
• fundmental alpha signals have lower trading turnover and lower stock coverage
• fundmental information tends to be reflected in stock prices over a relatively longer period of time

8. Equity Price and Volume

• the market is already close to , but never at, full efficiency
• The market is also evolving, the old models decay as the new ones emerge

9. Turnover

• an alpha with high excess returns and low variability consistently predicts future returns over a given time period
• turnover: the total value traded divided by the total value held

10. Backtest - Signal or Overfitting

• Three different simulation methods:

  • monte carlo simulation
  • pricing model
  • Explanation model

• one should not invest capital solely base on basktest simulation results

  • current market is not the same as historical period
  • Simulation assumption may not be realistic
  • possible forward-looking bias
  • overfitting(some time good simulation results that can be just random error or noise)

• Multiple technologies have been proposed to reduce overfitting risks

  • 10-fold cross validation
  • regularization
  • prior probability

• overfitting is easy

• correlation can cheat

• Financial markets have memory effect

• academic papers have bias

• A higer acceptance rule is needed

• How to avoid overfitting

  • out-of-sample test
  • Increasing in sample requirement
  • make model elegant
  • parameters and operations
  • Number of trials
  • Artifical data
  • Dynamic models

11. Alpha and Risk Factors

• In our practice, a good alpha tends to yield a higher Sharpe ratio after risk futralization, even though per-dollar return might be reduced in some cases

12. The relationship between alpha and portfolio risk

• we need to know the level of risk and return we started with in order to determine whether the alpha adds value

• Evety source of return derives form a source of risk, and so any portfolio can be viewed as a combination of risks, with weights on each risk according to the manager’s skill, perferences, and constraints.

• generating alpha means

  • search for sources of risk and return that are not known to others and
  • different from the ones that the manager is using already

• dilute exposure to risk sources that are widely konwn by other market participants

• A manager’s alpha comes from a combination of two sources:

  • factor timeing
  • tapping into risk sources unknowning to the rest of the market participants

• evaluating alpha’s performance

  • it is truly different? to diversify risk, the alpha must be orthohonal to widely-known factors, and to other alphas uesd in portfolio.
  • Who doen’t konw it? Attactive risk-adjusted retuens come mostly from ideas not known to other market participants.
  • Dose it make sense?
  • can it scale? Alphas that rely on mostly on factor timing are more difficult to execute, not least because their success depends to a greater extent on the cooperation of other market participants.
  • can it last? A good alpha should be non‐intuitive (preferably unappealing)
    to a fi rst‐level thinker, but intuitive and logical to a secondlevel thinker

13. Risk and Drawdowns

• measurement of risk of an alpha can simply be done by the level of volatility of its returns and the expected value at risk
  • Position concentration in a particular security or a group of securities
  • some alphas and alpha techniques are very commonly konon and are used by many quantitative researchers and portfolio managers around the world. become a source of risk.
• risk of and alpha or portfolio can be reduced by diversification
• drawdown is the percentage loss of an alpha from its peak.
  • the largest drawdown the alpha has through its history
  • the duration of the largest drawdown
  • attention should be normally be paid to avoiding largest drawdowns

14. Data and Alpha Design

• basic data: stock price, volume of a security
• how we find data for alpha
  • people always perfer good performance and low correlated alphas.
  • sometimes alpha sognals from the same dataset can be still highly correlated.
  • by using new data, we may achieve both performance improvement and diversification
• data from literature
  • the less well knowrn the data, the more valueable it can be
• data from vendors
• data validation
  • check the data usability(correct timestamp…, data delivery timeline, survivial bias)
• understand the data before using it

15. Statistical Arbitrage, Overfitting, and Alpha Diversity

• the key underlying assumption of statistical arbitrage is that the prices of financial instruments are driven by someconsistent rules
• based on a fake rule, an alpha may appear to be statistically significant in the data history, but then disappears in the future and never shows up again
• it’s more imprtant to understand the underlying price-driving rule, and use suitable implementations to take advantage of the rule to turn it into alphas.

16. Techniques for umproving the robustness of alphas

• The main goal of alpha research is to predict and outperform the market

  • Invariance under modification of traded universe
  • Robustness to extreme market conditions

• Simple methods for rubustness improvement

  • ordering methods
    • Ranking
    • quantiles approximation
  • approximation to normal distribution
    • fisher transform formula:
    • Z-socring: $F(r)=(r-mean(r))/stdev(r)$
  • Limiting methods:
    • Truncation: limit each stock to be within max percent of total position
    • winsorizing:limiting extreme values in the statistical data to reduce the effect of possibly spurious outliers

17 Alphas from Automated Search

• three components in an automated search
  • input data
  • search algorithm
  • signal testing
• make the input data ratio-like
  • compare the current value of a variable to its historical values
  • compare a variable with a similar variable of the same category
• input data should not come from too many categories (overfitting)
• it is not true that the longer the testing preiod the better
• sensitivity tests and significance tests are important

18. Algorithms and Special Techniques in Alpha Research

• Boosting (Adaboost)
• Digital Filtering(attenuate or amplify centain characteristics)
  • moving average
  • decompose the time series into trend and cycle compoents
• Feature extraction
  • PCA(finding patterns in data of high dimension)

PART III EXTENDED TOPICS

19. Impact of News and Social Media on Stock Returns

• News analytics and news sentiment are widely used by both buy-side and sell-side institutions in alpha generation, trading, and risk management
• less novelty news usually has smaller impact on the market
• some other news about industry or macro-economics usually has lower revelance to individual stocks
• different categories of news may have different response times on the market. Some categories have longer effects on company valuations, while other categories can cause short-term price fluctuations.
• Social media is a hot area in quant research

20 Stock Returns information from the stock options market

• The equity options market carries a lot of useful information for predicting stock returns

21 Introduction to Momentum Alphas

• momentum alphas are an important group and a broad category of alphas
• why momentum alphas work: A well-accepted theory is investors’ underreaction to new information
• factor momentun.Alphas that are based on factor regressions all assume that the factors’ returns have momentum effect
• group momentum. offen related to a phenomenon called co-movement
• momentum effects are related with inefficiency

22 Financial Statement Analysis

• Financial statement analysis attempts to systematically measure the effect of factors taken from the eaening statements and determine their ability to predict future returns.
• some well-known factors constructed from the balance sheet that positively correlated with future returns

23 Institutional Research 101

• Academic research on financial markets
  • freely avaliable academic papers on financial markets
  • be able to get inspiration, not just for trading ideas, but also for new methodologies
  • following davate may be useful if you are trying to figure out how to improve upon an existing framework or would like to know what most people say about its strengths and weaknesses
• Analysis research
  • Sell‐side analysts can perform analyses that are extremely costly, sophisticated, and time consuming, and they naturally want to provide first access to valued clients
  • Far more important than any specific “buy” or “sell” recommendation is the analyst’s thpught process
  • Analysts have detailed industry knowledge
  • analystes research can provide valid trading signals

24 Introduction to Futures Trading

• Futures trading includes equity indices, commodities, currencies, and bonds

25 Alpha on Currency Forwards and Futures

PART IV NEW Horizon-WebSim

• the market is always close to 100% efficiency but is never at 100%. we seek to arbitrage profit from this inefficiency

• Technical indicators are used for analyzing short-term price movements.

  • Relative Strength Index
  • Money Flow Index
  • MACD
  • Bollinger Bands

• Neutralization is an operation in which the raw alpha values are put into various groups, followed normalization.

  • alpha = alpha - mean(alpha)

• A good alpha would have comsistently increasing PnL and high Annual Return, Sharpe Ratio, % Profitable Days, and Profit per Dollar Trader, low Drawdown and Turnover. More importantly, it should’t have high fluctuations in the cumulative profit graph.

• academic/professional papers that can be used as a source of alpha ideas

PART V A FINAL WORLD

• The seven habits of Highly successful quants
  • willing to put in the extra effort
  • Always make sensible changes, don’t just try to fit the model to the data. Successfil quants will try to figure out the reasons for the suboptimal results, and try to make refinements that make sense.
  • Eager to experiment with new ideas
  • Do value-added work. Successful quants put in more effort in coming up with alphas based on news ideas.
  • Have a strong sense of urgency
  • Form synergistic terms
  • set high targets