Presentation is loading. Please wait.

Presentation is loading. Please wait.

Information Analysis. Introduction How can we judge the information content of an idea? More concretely: –Is this a signal we can profitably deploy.

Similar presentations


Presentation on theme: "Information Analysis. Introduction How can we judge the information content of an idea? More concretely: –Is this a signal we can profitably deploy."— Presentation transcript:

1 Information Analysis

2

3 Introduction How can we judge the information content of an idea? More concretely: –Is this a signal we can profitably deploy on paper? –Is this a signal we can profitably deploy in practice?

4 Outline Basic information analysis The information horizon Backtesting Fooling ourselves

5 Basic Information Analysis Can we profitably deploy this signal on paper? Two-step analysis: –Turn information (signal) into portfolios –Analyze portfolio performance Apply the scientific method –Hypothesis testing –Controls –Statistics

6 Basics Minimum Requirements –A process Assumption –Historical analysis will provide some useful insight into future performance. Observation –Running information analysis is easy. –Finding valuable information is hard.

7 Step 1: Information into Portfolios Keep in mind an example as we go through this. –B/P ratios We have a historical record of the signal. –Monthly B/P ratios back 10 years –What universe of stocks? How do we turn signals into historical portfolios?

8 Signals into Portfolios Choices here limited only by creativity. N-tile analysis: –Rank stocks by signal. –Place first (1/N) in N-tile 1, etc. –Even this simple approach offers many choices. Factor portfolios –Controlled experiments

9 N-tile Analysis Choices What is N? What universe of stocks? Do we divide stock universe –by number? –by cap? How do we weight stocks in each N-tile portfolio? –Equal-weight –Cap-weight –Something else?

10 B/P Quintiles: 1/88-12/92

11 Factor Portfolios Optimally construct portfolios Signal exposure one standard deviation above universe mean. Zero exposure to control variables Minimum risk.

12 Factor Portfolios Convert signal to “alpha” Build optimal portfolio

13 B/P Factor Portfolio: 1/88-12/92

14 Control Variables Other factors potentially related to returns: –The market (beta) –Industries –Investment themes (e.g. size, value, momentum…)

15 Alternative Approach Regression-based The coefficient f a is the return to a factor portfolio. Under what circumstances do these two approaches give the same answer?

16 Problem with Factor Portfolios They are non-investible. They take positions in every stock. They have long and short positions. We build them without regard for transactions costs. –So they are high turnover. Information analysis is the beginning of our analysis, not the last step.

17 Step 2: Performance Analysis We have now built historical portfolios for every period. –e.g. monthly B/P-based portfolios. We can now look at the performance of those portfolios over time. –Cumulative return plots –Alpha and beta of portfolios –t-statistic of the alpha –Information Ratio –Number of up versus down months –Performance in up and down markets –Turnover

18 Performance Analysis Key assumption: –Our signal is generating the performance. –We have controlled for all other relevant factors. We can examine such questions in detail

19 Holding and Return vs. Size

20 The Information Horizon The value of our information decays over time. –Information can grow stale. The Information Ratio will decay as information ages. The half-life of the information can quantify that effect. Longer horizons are typically desirable.

21 Measuring the Information Horizon Use time t information to build portfolios. Invest at time t+s. Monitor cumulative performance as a function of s. Monitor Information Ratios as a function of s. See examples

22 Cumulative Ret vs. Horizon (I)

23 Cumulative Ret vs. Horizon (II)

24 Cumulative Ret vs. Horizon (III)

25 Horizon IR (I)

26 Horizon IR (II)

27 Horizon IR (III)

28 Special Topic: Event Studies Particular form of information analysis. So far we have considered cross-sectional approaches. But what if we wish to study events which occur at different times for different assets? –Earnings announcements. –Change in CEO –Etc.

29 Event Studies The event occurs at t=0, no matter what the calendar date. We need to carefully control how we aggregate post-event returns, given that they involve different stocks at different times. Solution: we control for risk factors and risk levels.

30 Event Study Returns Look at standardized outcomes: We can combine these. We can regress against other controls:

31 -80-60-40-20020406080 1.12 1.08 1.04 1.00 0.96 0.92 0.88 Double Plus Universe Double Minus Number of days (earnings report = 0) Reaction time: weeks Cumulative value Source: R. Butman. DAIS Group. Example: Earnings Surprise Earnings surprise effects then (1983-89)...

32 Number of days (earnings report = 0) Reaction time: minutes to hours Cumulative value Source: R. Butman. DAIS Group. Evolution in markets … and earnings surprise effects now (1995-98) -80-60-40-20020406080 1.12 1.08 1.04 1.00 0.96 0.92 0.88

33 Backtesting More realistic version of information analysis. Investible portfolios. Account for constraints (like long-only) and transactions costs. These tend to lower performance relative to that observed in information analysis.

34 Backtesting, cont. Often backtests consider not only just lone signal performance, but the improvement in performance from adding new idea to existing alpha forecasts. Ultimately we will judge a signal based on whether it is: –Sensible –Predictive –Consistent –Additive

35 Fooling Ourselves Why are we so often disappointed by signal ideas that look great in backtests? We fool ourselves by interpreting statistical confidence very narrowly. Doesn’t a t-stat>2 imply 95% confident? –How many backtests should you need to run on independent data to have a 50% probability of observing at least one panel of random data exhibit a t-stat>2?

36 Four Steps for Backtesting Integrity Intuition Restraint Sensibility Out-of-sample Testing

37 Cautionary Tale Norman Bloom: World’s greatest dataminer.


Download ppt "Information Analysis. Introduction How can we judge the information content of an idea? More concretely: –Is this a signal we can profitably deploy."

Similar presentations


Ads by Google