DB European Quant Strategy - QM - Are Insiders Alpha Generators 20120926

Equity Quantitative Strategy Europe

26 September 2012

Quantitative Musing Is Insider Trading Information Valuable?

Are Insiders Alpha Generators? We investigate whether information contained in insider trades is valuable to investors. We find a number of metrics to be helpful in detecting which trades can add value for investors, and suggest a simple portfolio implementation to generate abnormal returns.

Quantitative Musing

Are Insiders Alpha Generators? Our monthly report for the quant and the non-quant investor alike

Deutsche Bank AG/London

Note to U.S. investors: US regulators have not approved most foreign listed stock index futures and options for US investors. Eligible investors may be able to get exposure through over-the-counter products. Deutsche Bank does and seeks to do business with companies covered in its research reports. Thus, investors should be aware that the firm may have a conflict of interest that could affect the objectivity of this report. Investors should consider this report as only a single factor in making their investment decision. DISCLOSURES AND ANALYST CERTIFICATIONS ARE LOCATED IN APPENDIX 1.MICA(P) 072/04/2012.

Glo

bal

Mar

kets

Res

earc

h

Qu

anti

tati

ve

Str

ateg

y

Team Contacts Jean-Robert Avettand-Fenoel [email protected] Spyros Mesomeris [email protected] Marco Salvini [email protected] Yiyi Wang [email protected] (+44) 20 754-71684

Is Insider Trading Information Valuable? In this report, we investigate whether information contained in insider trades is valuable to investors. We find that not all trades are equally informative. Instead, we uncover a number of metrics which help us to detect high conviction trades. A simple systematic portfolio implementation of directors’ high conviction trades generates significantly positive abnormal returns.

The Characteristics of Insider Deals Before dwelling on the performance of insider deals, we provide a description of the insider trading activity in the European market, more precisely in the S&P BMI Europe universe. After reaching all time highs in 2007-2008, the number of trades executed on a monthly basis has now normalized. While large caps have seen a decrease in insider trading relative to small caps, no particular trend or imbalance can be observed at the sector and country level.

Uncovering Profitable Director Deals Unconditional backtest analysis of insider trade performance, either via event studies or more practical portfolio construction exercises, show that insiders outperform in the long term, but “following” all trades might be costly and not sufficiently rewarding. For that reason, we investigate a number of metrics which prove useful in selecting directors’ trades. These characteristics revolve around four main ideas, namely: the limited attention hypothesis, insider trade dynamics, information asymmetry, and confirmatory/contrarian trading. Event studies show that, over both the short- and the long-term, the metrics we study can help differentiate between future outperformers and underperformers. We further use a machine learning algorithm called FOREST to assess whether our findings appear robust once the threshold used to split the data by each metric is not pre-determined. Finally, we look at whether diversification benefits accrue to combining the various metrics together.

Outperforming with Insider Information Directors’ deals cannot be implemented as traditional alpha signals, due to their event-driven nature. Moreover, only a small proportion of companies have insider trading activity in any given month. Therefore, we propose a simple portfolio implementation consisting of a satellite portfolio that tracks insider trades alongside a core benchmark portfolio. In particular, the weight allocated to the insider portfolio can be fine-tuned to target a given tracking error to the benchmark. Being selective on insider trades allows us to generate twice the Information Ratio compared to a portfolio that does not filter trades by the degree of conviction. We further show our trade selection process is not the consequence of randomness, and that the final portfolio has no unintended exposure to traditional risk factors.

26 September 2012 Quantitative Musing


Table of Contents

Is Insider Trading Information Valuable? ........................................ 3 The Basics of Directors Dealings .............................................................................................. 3 Not All Information is Valuable .................................................................................................. 3 Profiting From Director Deals ................................................................................................... 4

The Characteristics of Insider Deals ................................................ 5 A European Database for Director Dealings ............................................................................. 5 Unconditional Backtests of Director Deals ................................................................................ 9

Uncovering Profitable Director Deals ............................................ 11 Univariate Signal Tests ............................................................................................................ 11 Assessing Selection Criteria via Classification ........................................................................ 23

Outperforming with Insider Information ....................................... 26 Satellite Insider Portfolios ....................................................................................................... 26 Robustness of the Insider Strategy Returns ........................................................................... 28

References ........................................................................................ 30



Is Insider Trading Information Valuable? The Basics of Directors Dealings

Insider trading is sometimes associated with unlawful behavior; for instance, directors trading with prior knowledge of confidential and private information, and trying to hide these trades. This is very different from the kind of insider trading we investigate in this paper. Indeed, for an insider trade to be legal, the director must report it to the local regulator (such as the SEC for the US) or disclose it publicly (for instance in the “Directors Dealings” section of the Financial Times). From that point onwards, the director with insider knowledge will have shared the information she possess with the market, her trade potentially acting as a “Buy” or a “Sell” signal, accordingly.

Nevertheless, such trades don’t always serve the purpose of signaling. As insiders’ shares are often obtained in the form of compensation awards, the impetus for selling could be monetization of the insider’s reward, rather than the expression of a negative view on the future prospects of the company. On the contrary, share purchases could prove more meaningful, potentially reflecting the real intention of participating in share price performance. The ultimate question is therefore how consequential each insider trade is, and how much private information each trade conveys to the market.

A significant amount of academic literature has explored this topic since the late 1970s. Even though, in general, the conclusion is that insiders are indeed better informed and are able to generate positive abnormal returns through their trades, exploiting this information from a public investor point of view can prove tricky. As we show later in the report, we are not able to construct a profitable strategy if we do not discriminate, a priori, between insider trades, as returns are not high enough to overcome the cost of trading. Consequently, a more recent strand of the literature has explored segregating insider trades into profitable and unprofitable ones.

Not All Information is Valuable

To start with, it is worth pointing out that trades reported by directors can be of several different types. Obviously, we observe a significant amount of “regular” market transactions, but we also observe trades of less significance such as exercise of employee options, IPO- related transactions, dividend reinvestments, or simply a transfer of shares to the director’s spouse or dependants. There is a natural pecking order regarding the signaling significance of insider trades, which calls for a first stage of filtering to retain mostly “regular” market transactions, which should more clearly serve as higher quality signals.

Thereafter, we should expect a certain degree of divergence among “regular” market trades. First of all, some directors might be better informed than others, and/or be better at disseminating the information they possess to the market1. A related topic is the position held and rank of the respective director in the company, in the sense that a trade carried out by the CEO or CFO, for instance, would not be expected, a priori, to have the same impact compared to a trade done by a lower level executive. Similarly, the size of the transaction and

1 Kyle (1985) in particular investigated the relationship between informed traders and liquidity in a sequential auction to study the optimal trading strategy to incorporate private information into prices gradually while maximizing expected profits.

Sometimes, insider trading

is associated with illegal and

criminal behavior. In this

report we investigate lawful

insider trading, i.e. trades

reported to the local

regulator.

The impetus for selling

could be monetization of the

insider’s reward, rather than

the expression of a negative

view on the future prospects

of the company.

However, share purchases

could potentially reflect a

real intention of

participating in share price

performance.

There is a natural pecking

order regarding the

signaling significance of

insider trades, which calls

for some filtering to retain

mostly “regular” market

transactions.

Additionally, we should

expect a certain degree of

divergence among “regular”

market trades.



the number of insiders trading at any one time could signal a higher or lower degree of conviction, and be interpreted as such by the market.

Nevertheless, the aforementioned metrics that could help us differentiate between insider trades reflect insider- and transaction-specific characteristics. Academic papers have also envisaged analyzing company-related metrics. In a similar fashion to our share buybacks research (see Salvini et al. (2012)), a potential source of differentiation regarding the significance of insider trades could come from the asymmetry of information between insiders, and the market. Namely, if the asymmetry is almost non-existent (high analyst coverage, large caps well understood by their investors), the incremental information brought to the market by an insider transaction could be rather low, while in a company where asymmetry of information is preponderant (for instance in high R&D intensive companies), an insider transaction could reveal a lot of useful information to investors.

Finally, market perception could also impact the significance of insider trades. For instance, insiders might want to trade “against” the market to try and restore confidence in their company. This could be the case, for instance, when a company reports its earnings and the market reacts negatively to the announcement, leading to share price underperformance: An insider purchase following such an announcement could signal a belief that better prospects lie ahead for the company in question, reassuring the market, while at the same time profiting from the temporary drop in the share price. Similarly, an insider buying after a positive surprise would serve as a confirmatory signal that the company is in a good shape.

Another way to detect market perception/sentiment towards a particular stock would be to look into the securities lending market. As we have observed in prior research (see, for instance, Cahan et al. (2011), Avettand-Fenoel et al. (2012)), stocks which are highly shorted tend to subsequently underperform the market. With this in mind, insiders buying shares which are heavily shorted would send a positive conviction signal for the stock, whilst insider selling in the presence of significant short interest might suggest a longer stretch of difficult times ahead for the respective company.

We will explore in-depth whether and how all the factors mentioned above determine/affect stock returns following insider trades, and how they interact with insider trades to impact the quality of the signal.

Profiting From Director Deals

It is important to mention that implementing traditional quant signals based on insider trading is not an easy task. Indeed, any strategies utilizing director deals should rather be classified as event-driven strategies, since not all companies have insiders carrying out trades. This means that in a universe of 1000 stocks, only a certain percentage of stocks would generate signals at any point in time, which makes it hard to create diversified baskets exploiting this “premium” cross-sectionally.

In earlier papers we have proposed several ways to implement systematically such event-driven strategies (see for instance our paper on Earnings Surprises,2 or our paper on M&A News3). Here, we will employ another simple methodology which aims to leverage the information contained in insider trades to run a satellite portfolio alongside a core benchmark index. In such a portfolio set-up, the weight allocated to the insider strategy can be tuned to target a given tracking error.

2 See Mesomeris, S., et al., 2011, « Quantitative Musing – Earnings Surprises: It’s all about Beat or Miss », Deutsche Bank Equity Quantitative Strategy, 8th June 2011. 3 See Avettand-Fenoel, J.-R., et al., 2011, « Quantitative Musing - Targeting M&A News », Deutsche Bank Equity Quantitative Strategy, 12th October 2011.

Among the various metrics

used to discriminate

between insider trades,

some reflect insider- or

transaction-specific

characteristics, while others

relate to the company itself.

Market perception could

also impact the significance

of insider trades, as insiders

might want to trade

“against” the market to try

and restore confidence in

their company, or trade to

“confirm” positive

prospects.

Investment strategies

relying on insider trading are

mainly event-driven. We will

employ a simple

methodology which “runs”

a satellite portfolio along a

core benchmark index.



The Characteristics of Insider Deals A European Database for Director Dealings

As we have mentioned in the Introduction, insider trades are made legal when directors report them to their respective local regulators. This particular fact has actually been a non-negligible issue for quants wishing to exploit such information. Considering the various regulations across different European countries, director trades have been reported historically in an inconsistent manner, making it hard to build a database with comparable information for all European insider transactions. It’s not until 2004 that a European directive has been put in place to harmonize and standardize the reporting framework. But even so, assembling a pan-European insider dealing database would still be a cumbersome task.

Fortunately enough, a database provider called 2iQ, is already doing all the necessary work of collecting, cleaning, and storing insider transactions data. Covering more than 20 European countries from the early 2000s onwards (the starting date varies for each country), the data is presented in a harmonized form, which is a pre-condition for using a database in a systematic strategy. Notwithstanding possible reporting errors, manual checks are carried out upstream, for instance regarding an insider’s position in the company. Then, for each single trade, a multiplicity of items is provided, such as the transaction type, the insider’s name and level in the hierarchy, the number of shares traded, the total value of the transaction, etc. (see Figure 1). Furthermore, 2iQ also aggregates data points to create additional informative database items, such as accumulated volume of insider purchases/sales over a recent period, or the number of insiders buying/selling the same stock, etc.

Figure 1: Snapshot from the 2iQ Database

Company Name SEDOL Insider ID Insider Name

Insider Relation Insider Level

Transaction Type

TransactionCount

Shares Input Date Price Value Curr.

WENDEL 7390113 154550 Gautier; Bernard

Executive Board A Sell 2 1306 30/03/2011 74.06 97007 EUR

EFG EUROBANK ERGASIAS SA

5654316 180826 Marinos; Georgios

General Manager B Buy 1 3000 17/12/2008 5.763 17289 EUR

SANDVIK AB B1VQ252 195838 Lundberg; Fredrik

Director/Beneficial Owner

C Sell 1 2mn 10/12/2008 47.7 954mn SEK

GL EVENTS 7154104 352940 Weber; Caroline

Non-Executive Director

C Buy 1 100 28/09/2011 19.11 1911 EUR

KLOCKNER & CO SE B170DQ6 241659 Ruehl; Gisbert

Chairman of the Board

A Buy 1 6426 14/05/2010 17.3878 111734 EUR

SIRTI SPA B18P5C0 159526 Lualdi; Ambrogio

CEO A Sell 5 140k 26/10/2006 2.774 396069 EUR

MORGAN CRUCIBLE CO PLC

0602729 136987 Knowlton; Warren D.

CEO A Buy 1 100k 18/03/2003 45 4.5mn GBX

ENGR-INGEGNERIA INFORMATICA

4820453 150973 Cinaglia; Michele

Non-Executive Vice Chairman

C Buy 1 500 26/01/2008 22.75 11375 EUR

KOMPLETT ASA B00KX66 155550 Hagen; Stein Erik

Director C Buy 2 2.1mn 13/12/2007 122 256.9mn NOK

SKANDINAVISKA ENSKILDA BANK

4813345 196198 Ovesen; Jesper

Director C Buy 1 480 07/08/2006 178.5 85680 SEK

Source: 2iQ Research, Factset, Deutsche Bank

Considering the various

regulations across different

European countries, director

trades have been reported

historically in an

inconsistent manner.

A database provider called

2iQ, is doing all the

necessary work of

collecting, cleaning, and

storing insider transactions

data, covering more than 20

European countries from the

early 2000s onwards.



In order to get a better understanding of insiders’ behavior, we start our analysis by showing aggregated trade statistics for our investment universe, the S&P BMI Europe. It’s worth pointing out that, in line with the academic literature, we first filter out a significant number of trades: we obviously keep only trades in equity shares and only regular buy and sell trades (i.e. no options exercise, or subscription to a capital increase). Transactions by lower level executives or “outsiders” like trusts and family holdings (classified as insiders because they are required to report their trades) are excluded from the sample. Finally, we also remove OTC transactions and penny stocks.

Number of Deals across Time To start with, in Figure 2 below we show the number of deals across time, which we further differentiate into purchases and sales. The two time series exhibit an intriguing evolution over time. While the number of insider deals increased moderately through to 2006, it jumped to all-time high levels during 2007-2008, before dropping and stabilizing at about 500 deals per month thereafter. This kind of pattern reminds us strongly of M&A activity, share buybacks, and, more generally, equity market volume patterns (although it is also influenced by the fact that the reporting start date varies among countries, with more countries being added in the database as time progresses).

The split between buy and sell trades has been far from a one for one ratio. While it is the case in a few months, the dominant trades through time are insider purchases. Interestingly, while we observe some spikes in the buy/sell ratio, we don’t see the opposite, i.e. a lot of sales and few purchases. Finally, Figure 3 tells us that companies with insider trades have, on average, 2 trades executed in a month, and among the whole S&P BMI universe, about 13% of all companies have insider trades. The aforementioned numbers suggest that, despite about 70,000 insider trades included in the database, this number is rather small to generate sufficient signals for the full universe of stocks all the time.

Figure 2: Number of Insider Deals per Month

Figure 3: Number of Companies with Insider Deals per

Month

Source: 2iQ Research, Factset, Deutsche Bank Source: 2iQ Research, Factset, Deutsche Bank

Our investment universe is

the S&P BMI Europe. We

obviously keep only trades

in equity shares, and only

regular buy and sell trades

After rising to all-time highs

in 2007-2008, the number of

deals per month is now

stabilizing at about 500.

We usually observe more

purchases than sales.

Companies have on average

2 insider trades each month.



Insider Deals Breakdown by Market Capitalization The following two charts (Figure 4 and Figure 5) display a breakdown of all purchases and all sales by market capitalization segment (Small, Mid, and Large Caps). It is interesting to note that the share of large cap insider purchasing activity has been constantly decreasing over time, whereas, on the contrary, small cap insider purchasing, whilst appearing to be pretty low at the beginning of the sample, has been on a rising trend.

On the other hand, there is no clear time-series variation in the proportion of insider-selling activity across market capitalization segments; that is, the proportion of sales attributed to each capitalization segment has remained rather constant through time. On an absolute level, large- and mid-capitalization stocks have accounted for the lion share of sales, as opposed to small-caps, and contrary to what we have observed for purchases.

Figure 4: Breakdown of Insider Purchases by Market

Cap Segment

Figure 5: Breakdown of Insider Sales by Market Cap

Segment


Insider Deals Breakdown by Sector Figure 6 exhibits each sector’s share of insider deals averaged over the whole data sample, as well as the sectors’ share in the benchmark, where shares are calculated in number of stocks. In fact, the pie chart depicting each sector’s share in the benchmark looks very similar to the ones showing purchases and sales by sector, suggesting that no undue buy or sell activity takes place in any sector. This was not so obvious on an ex-ante basis, as we would probably have expected a greater intensity of insider trades in sectors which are likely to be associated with more asymmetric information. We see here that this is not the case, at least from a transaction count point of view.

The share of large cap

insider purchasing activity

has been constantly

decreasing, while small cap

insider purchasing has been

on a rising trend. There is no

clear time-series variation in

the proportion of insider-

selling activity across

market capitalization

buckets.

No undue buy or sell activity

takes place in any sector as

the breakdowns are in-line

with the benchmark.



Figure 6: Average Share of Insider Trades by Sector


Insider Deals Breakdown by Country Our last slicing of the data, exhibited in Figure 7 and Figure 8, is by country (for clarity, we do not show the legend as more than 20 countries are included in either Figure; note that each color represents an individual country). In a similar fashion to the sectoral breakdown, no particular trend emerges.

Instead, what we observe is a clear illustration of the fact that we have variable start dates for different countries. At the very beginning of the time series (i.e. early 2003), we have only nine countries with insider trades: UK, Germany, France, Italy, Belgium, Austria, Netherlands, Sweden and Finland. Then, as time passes, more and more countries start to report insider transactions, and thus enter the database.

Figure 7: Breakdown of Insider Purchases by Country Figure 8: Breakdown of Insider Sales by Country


In a similar fashion to the

sectoral breakdown, no

particular trend emerges.

Instead, we observe a clear

illustration of the fact that

we have variable start dates

for different countries.



Unconditional Backtests of Director Deals

Before considering sorting trades into “convincing” and “less convincing” categories, we run a battery of unconditional backtests in order to understand whether insider trades, in general, add value to investors acting upon such trades. The first methodology used is an event study analysis, where we assume that we buy (sell) the stock the day after the insider purchase (sale) has been reported in the database. We then assess the performance of the trade relative to the index, i.e. the S&P BMI Europe, in our case. We have a 12-month horizon over which to assess trades, ranging from date T+1 to date T + 12 months, where T represents the insider transaction date.

Figure 9: Excess Return of Insider Purchases vs. S&P

BMI Europe from T+1 month to T+12 months

Figure 10: Excess Return of Insider Sales vs. S&P BMI

Europe from T+1 month to T+12 months


We can see in Figure 9 that buy transactions outperform the benchmark over the long term (by about 3.3% over 12 months), but this not the case for share sales, which only mildly underperform over the short term (see Figure 10). Using confidence intervals at the 97.5% significance level confirms that the underperformance is mild for share sales (-0.6% after 4 months) and not significantly different from zero after 8 months.

Furthermore, to render the backtest slightly more practical, we draw on a rather “naïve” strategy whereby we track insider trades by forming a portfolio each month which makes the same investment as the insiders, while the final portfolio is a combination of the benchmark index and the insider portfolio, the objective being to outperform the benchmark with a reasonable tracking error. Two parameters are driving the portfolio construction: the holding period (i.e. the number of months over which we hold the stocks), and the weight allocated to the insider strategy in the final portfolio.

We first run unconditional

backtests using an event

study, where we assume

that we buy (sell) the stock

the day after the insider

purchase (sale) has been

reported in the database.

Buy transactions outperform

over the long term, while

share sales only mildly

underperform over the short

term.

In this “naïve” strategy, the

final portfolio is a

combination of the

benchmark index and the

insider portfolio.



Figure 11: Information Ratio of the Active Insider Strategy, with Variations on the Weight allocated to the Insider

Strategy Versus the Benchmark, and on the Time (in Months) over which Insider Trades are Held

Transaction costs are assumed to be 20bps each way. Source: 2iQ Research, Factset, Deutsche Bank

Figure 11 above shows the Information Ratio (risk-adjusted excess return over the benchmark) of the backtest analysis where we vary the two parameters, Weight and Holding Period, one after each other. We first fix the holding period at the 1 month horizon, and vary the weight allocated to the Insider strategy (at 10%, 25%, 50%, and 100% respectively), and subsequently hold the weight constant at 25% and vary the holding period. Before analyzing the influence of each parameter on the result, we first confirm the previous findings, namely that following insider buying can be rewarding, while following insider selling is not a viable strategy as the information ratio is either flat or negative.

The results suggest that increasing the weight allocated to the insiders’ strategy doesn’t impact the performance of the strategy materially; although post-transaction costs the risk-adjusted return (IR after costs assuming 20bps each way as a fee) decreases mildly. Next, assuming we keep the allocation to the insiders’ strategy constant (at 25%), we investigate how varying the holding period influences the results. In both cases (buy and sell), longer holding periods are not beneficial to the strategy from a return point of view. However, longer holding periods result in lower turnover for the portfolio, whilst not eroding performance significantly; in fact, gross and net Information Ratios converge as the holding period stretches out.

The observations above lead us to reflect that profiting from insider trading is not a futile task. Without being selective on trades, results appear to be rather encouraging. Yet, the strategy is likely to suffer from high turnover, and looks less appealing after transaction costs are accounted for. In order to achieve superior risk-adjusted returns (net of transaction costs), common sense tells us we should be more selective on trades, focusing on the ones which are likely to be more convincing.

Two parameters come into

play in the strategy: the

weight allocated to the

insider portfolio versus the

benchmark and the holding

period of the insider

portfolio.

Longer holding periods are

not beneficial to the strategy

from a return point of view,

but result in lower turnover

for the portfolio. The risk-

adjusted return is thus not

adversely impacted.

In order to achieve superior

risk-adjusted returns, we

should focus on trades

which are likely to have

higher signaling value.



Uncovering Profitable Director Deals Univariate Signal Tests

The fact that the information content of insider trades is variable has been well documented in the academic literature. As we mentioned above, it seems intuitive that we first need to carefully select trades that are likely to signal higher conviction than others. Yet even among these, some insider trades may still carry no meaningful information, as suggested by Giamouridis et al. (2008) and Dardas (2011).

A significant number of metrics have been investigated in academia over the last few years in order to uncover which characteristics of director dealings can be useful for detecting profitable trades. While the usefulness of such metrics typically depends on the country, the universe, and the timeframe used for the investigation, we find a number of them to be universally applicable. In order to better communicate our findings, we employ event studies yet again, where trades are selected on the basis of a given metric. Importantly, to avoid any bias caused by potential outliers in our study, we carry out random sub-sampling of the trades in each category (forming 100 sub-samples), computing the event study again for each of the subsamples, and then re-aggregating the results to show the confidence intervals at the 97.5% significance level. Based on the results analyzed in the previous Section, hereafter - for the univariate signal tests -, we focus on purchases.

Size/Value/Momentum Before plunging into insider-specific metrics, we start by investigating typical characteristics of insider trades that are often included in risk and quantitative alpha models, such as Size, Value and Momentum. We simply define these factors by Market Cap (in EUR billion), Book-to-Price, and First-11M Price Momentum, respectively. Instead of applying the usual quintile sorting of stocks in the cross-section, we prefer to work only with two “buckets” given the small number of firms with insider trades on a monthly basis (often less than 20% of the universe).

The question we are trying to answer is whether splitting director purchases into cheap versus expensive stocks, small versus large and low versus high price momentum stocks can give us an edge in predicting future share price performance. In other words, is director buying significantly associated with any of the standard factor premia? In Figure 12, we observe, surprisingly, that factors such as Size and Value cannot be used to statistically distinguish between the longer-term performance of director deals (indeed, confidence intervals are clearly overlapping). A clearer pattern appears in the case of Momentum, but as we will see later, other factors can achieve a better distinction between trades.

The information content of

insider trades is variable.

A significant number of

metrics have been

investigated in academia

over the last few years in

order to uncover which

characteristics of director

dealings can be useful for

detecting profitable trades.

We start by investigating the

question using typical

characteristics that are often

included in risk and

quantitative alpha models,

such as Size, Value and

Momentum.

Surprisingly, factors such as

Size and Value cannot be

used to statistically

distinguish between the

longer-term performance of

director deals.



Figure 12: Excess Return of Insider Purchases Split Into Low and High Size/Value/Momentum


Insider Hierarchy One of the most intuitive attributes of director deals worth investigating is the position of the insider in the pecking order of the company. In particular, it seems natural to assume that the position a director occupies in the company’s hierarchy may affect, if not determine, the trade’s conviction. According to this hypothesis, a trade conducted by the Chairman or CEO should provide the most useful signal, followed perhaps by the CFO and other senior executives. To analyze this, we first take a look at three event study analyses isolating trades made by Chairmen, CEOs and CFOs, in Figure 13.

Figure 13: Excess Return of Insider Purchases Split by Insider Rank: Chairman, CEO, and CFO, vs. Other Insiders


It seems natural to assume

that the position a director

occupies in the company’s

hierarchy may affect, if not

determine, the trade’s

conviction.



The results in Figure 13 appear to contradict our expectations. Instead of observing a significant outperformance for Chairman and CEO trades, we rather observe the opposite pattern: Specifically, trades executed by company board Chairmen don’t generate high returns on an absolute basis, even starting to significantly underperform trades made by other directors four months after the trade date. Share purchases by CEOs tend to perform in line with other director trades for the first 6 months, followed by a significant underperformance thereafter. On the contrary, CFO purchases appear to outperform trades made by other directors both in the short- and in the long-term.

The previous results suggest that factors other than information hierarchy are at play here. Rather, as suggested by Jeng, Metrick, and Zeckhauser (1999), CEO trades don’t earn higher abnormal returns for the simple reason that they are under such scrutiny that they can’t easily place informed trades in the market, without drawing too much attention. Furthermore, the reversion of the CEO trade performance could also emphasize the argument that investors wishing to follow CEO trades would end up arbitraging the trade.

When we look at trades carried out by CFOs, instead of CEOs, we observe that abnormal returns are significantly higher at all horizons. A plausible explanation for this pattern might be that although CEO actions are widely scrutinized by the market, this is much less the case for CFOs and their trades. In order to substantiate the argument further, we follow Dardas and Guttler (2011) and split all the insider levels in three categories: A, B and C, where A corresponds to top insiders (such as CEO, Chairman, etc.), B corresponds to upper level management (executive committee, etc.) and level C corresponds to non executives such as the supervisory board. We can see from Figure 14, where the frequency of each insider level is plotted, that Level A insiders are the most frequent, followed by C and then B. Figure 15 shows us how trades performed by each level of insiders have performed against the benchmark.

Figure 14: Frequency of Insider Levels in ”Buy” Trades

Figure 15: Excess Return of Insider Purchases Split by

Insider Level


Interestingly, the results above seem to confirm the hypothesis that widely “followed” insider purchases (by level A Insiders) significantly lag behind, in performance terms, those conducted by less “famous” insiders (level B). The third level of directors, level C, which is non-executive and thus less involved in the day-to-day operations, seems to be less relevant.

Unexpectedly, Chairmen and

CEO trades underperform

purchases made by other

insiders. On the contrary,

CFO purchases outperform.

CEO trades don’t earn

higher abnormal returns as

the forces of arbitrage

seemingly act quickly, given

the attention such trades

draw upon themselves.

To corroborate the results,

we follow Dardas and

Guttler (2011) and split all

the insider levels in three

categories A, B and C.

The results confirm that less

“followed” insider (level B)

purchases earn significantly

higher excess returns than

those conducted by level A

insiders.



Transaction Size The second metric we looked at was the value, or size, of the trade. Intuitively, we expect that high conviction insiders would trade a significant amount of shares, while others would not put much money at risk. Seyhun (1986) found, for instance, that higher transaction values were usually associated with higher abnormal returns (thus suggesting that the information contained in the trade was proportional to its value), while Barclay and Warner (1993) argue that informed investors are better off trading in medium sizes in order to avoid revealing too much information to the market. This argument would be consistent with Kyle (1985), who demonstrates that informed traders are better off trading moderately in order to control how much information they reveal to the market. In order to understand better how trades of various sizes react, we look at an event study in Figure 16 which splits trades into two categories: large trades, which are larger than the median trade made during the previous month, and small trades which are smaller in size than the median trade made during the previous month.

Figure 16: Excess Return of Insider Purchases Split by Transaction Size


The results in Figure 16 seem to favor the latter hypothesis, namely that informed and profitable trades are those that are smaller in size. This observation actually ties in well with that made earlier regarding insider rank: market participants willing to exploit insider information would first look for the obvious (i.e. trades by Chairmen/CEOs, or trades large in size), neglecting the smaller trades made by upper level management. In line with the limited attention hypothesis, these trades consequently exhibit higher expected returns.

Nevertheless, we note that the absolute value of a trade may be a flawed measure of conviction in the first place. Indeed, we could investigate the impact of trade size on a relative basis, zooming, for instance, on the number of shares traded relative to the initial holdings. A director doubling her holdings potentially signals a different degree of conviction relative to a director increasing her holdings by a mere percentage point. The former looks more strategic than the latter. Additionally, we could also compare the trade size to the total number of shares outstanding, potentially reflecting a significant increase in the director’s overall stake in the company.

The academic literature

finds conflicting results as to

whether large or small

transactions earn higher

returns.

The results seem to favor

the hypothesis that

informed and profitable

trades are the ones that are

smaller in size.

Nevertheless, the absolute

value of a trade may be a

flawed measure of

conviction in the first place.

The size of a trade relative

to a director’s current

holdings may present a

stronger signal.



Figure 17: Event Study with Trade Size as a Percentage

of Director’s Holdings

Figure 18: Event Study with Trade Size as a Percentage

of the Number of Shares Outstanding


In Figure 17, it can be observed that 12-month forward excess returns increase in a rather monotonic fashion in the trade size (as a percentage of the director’s holdings). While the bulk of all trades usually involve small changes in directors’ holdings, a non-negligible number of them lead to significant changes. One could further argue that small trades (relative to holdings) would constitute some kind of “noise” trading: a marginal change in holdings is unlikely to materially change the insiders’ wealth in the longer term. However, significant share purchases could reflect a more strategic view on the part of the director, which might explain their superior performances.

On the right hand side, Figure 18 depicts the results of a similar analysis carried out with trade size as a percentage of the total shares outstanding. These are less clear-cut. Indeed, the first four categories where trades represent less than 0.5% of the total shares outstanding seem indistinguishable. However, large trades of 0.5% of shares outstanding and above stand out, exhibiting substantially higher excess returns on a 12-month horizon.

Net Insider Volume Ratio (NIVR) and Net Insider Count Ratio (NICR) The two previous metrics, Insider Level and Transaction Size, have both aimed at inferring whether a trade embeds superior information or not simply on the basis of what has been reported for a specific transaction. In the spirit of Lakonishok and Lee (2001), one could also explore the evolution of insider trading activity over a certain period. More precisely, we employ a metric called the Net Insider Volume Ratio (NIVR), which aggregates insider purchases and sales over the previous 6 months, and then calculates the ratio of the difference (purchase volume minus sales volume) divided by the sum (purchase volume plus sales volume). Such an indicator would range from -1 (corresponding to pure sales over the last 6 months) to +1 (only purchases). In the middle of the spectrum, a NIVR of 0 indicates a neutral stance, as there are as many purchases as there are sales.

Trade size as a percentage

of holdings is a useful

criterion as small trades

could constitute “noise”

trading, while large trades

could be reflecting a more

strategic view on the part of

the insider.

Only large trades of 0.5% of

the total shares outstanding

and above stand out,

exhibiting higher excess

returns on a 12-month

horizon.

In the spirit of Lakonishok

and Lee (2001), we also

examine the dynamics of

insider trading, using the

Net Insider Volume Ratio

(NIVR) over the last 6

months.



Figure 19: NIVR Distribution among Insider Purchases Figure 20: Event Study based on NIVR


Interestingly, we observe in Figure 19 that a disproportionate number of purchases have a NIVR of 1. Actually, in our sample, we almost have the same number of trades with a NIVR of 1 as trades with a NIVR strictly less than 1. Talking about conviction, a NIVR less than 1 would reflect either uncertainty or a change in conviction. This is confirmed in Figure 20 where we compare the performance of stocks purchased by insiders with a NIVR negative, positive, or equal to 1. If only insiders’ purchases were made over the last 6 months, a new insider purchase should then be considered as credible, as these trades usually outperform significantly other trades where sales have been involved. Indeed, if sales volume is also observed in the past 6 months (thus leading to a NIVR strictly below 1), forward returns are lower upon a new insider purchase. In other words, it seems that consecutive director purchases impart a certain degree of positive price momentum to the underlying stock, which may also explain, at least in part, the findings in Figure 11.

Additionally, apart from analyzing the evolution of insider trading behavior from a volume point of view, it is rather simpler to look at the aggregated behavior of the insiders in each company. Intuitively, if various insiders from the same company buy their company’s stock, this should measure up to be a more convincing signal relative to an isolated insider buying the stock. Even worse, a situation where an insider is buying while all her colleagues are selling would almost certainly not look that positive in retrospect. Consequently, we build a metric called the Net Insider Count Ratio (NICR) which in a similar fashion to the NIVR, uses the number of insiders buying or selling over the last 6 months, as opposed to transaction volumes. We should thus be able to detect “clusters” of insiders with a positive/negative stance on their company.

If only insiders’ purchases

were made over the last 6

months, a new insider

purchase should then be

considered as credible, as

these trades typically

outperform significantly

going forward.

Instead of analyzing the

evolution of insider trading

activity from a volume point

of view, we also calculate

the difference between the

number of insiders buying

and selling over the last 6

month divided by the total

number of insiders trading,

which we call the Net

Insider Count Ratio.



Figure 21: Event Study based on the NICR


Results in Figure 21 using the NICR follow somewhat closely the ones obtained with the NIVR. This suggests that no matter if we look at insider volume or insider count, what is important is that insiders in a same company do not express diverging views on the stock.

Trade Isolation Apart from exploring insider trading momentum via the NIVR/NICR, another way to investigate this could be trough the trading frequency of a given insider. More precisely, Cicero et al. (2012) suggest splitting trades into two categories: isolated trades and sequenced trades. Isolated trades happen in a month following a prior month when the insider didn’t trade at all, while sequenced trades correspond to purchases/sales made over a number of successive months by the same insider. Information disseminated less regularly is less expected by the market and may lead to underreaction towards the release of an “unscheduled” piece of information as conveyed through an isolated trade. Indeed, we might expect that sequenced trades do not carry much information, since the market already knows the position of the insider from previous months, and grows to expect her trading pattern. This would mean that most of the information is conveyed in the first trade, and the market incorporates less and less information as trades are made in sequence. Furthermore, in the case of strong information asymmetry, insiders would potentially prefer to trade sporadically so that investors wouldn’t pay too much attention.

Results using the NICR

follow somewhat closely

those obtained with the

NIVR.

Isolated trades happen in a

month following a prior

month when the insider

didn’t trade at all, while

sequenced trades

correspond to

purchases/sales made over

a number of successive

months by the same insider.



Figure 22: Excess Return of Insider Purchases Split into Isolated and Sequenced

Trades


We plot in Figure 22 the event study where trades are split into the two aforementioned categories: isolated trades (i.e. no trade made by the insider in the prior month) and sequenced trades (i.e. trades by the insider in several successive prior months). It appears quite clearly that, from T = 3 months up to T = 12 months, excess returns differ significantly between isolated and sequenced trades, and increasingly so. Isolated trades conveying new information are rewarded, while sequenced trades, which don’t provide particularly new information to the market, generate returns not significantly different from zero over the long term.

R&D Intensity It is worth noting that insider trading shares a common characteristic with share buybacks, which we investigated in Salvini et al. (2012). Indeed, they are both actions taken by a company’s top executives that can reveal to the market information about the company itself. In our report on buybacks, we were also aiming at detecting share buybacks that convey higher quality information, and among others, we found R&D Intensity to be a particularly useful metric. R&D Intensity, defined as R&D expenses divided by total assets, is often used as a proxy for information asymmetry, as R&D-intensive firms tend to disclose little information regarding their innovations until they are released to the market. Aboody and Lev (2000) already explored the relationship between insider gains and R&D intensity, and indeed found that it was a clear determinant of information asymmetry and of the profitability of insider trades.

Isolated trades, which

conveying new information

to the market are rewarded,

while sequenced trades

generate returns not

significantly different from

zero over the long term.

Aboody and Lev (2000)

already explored the

relationship between insider

gains and R&D intensity and

indeed found it to be a clear

determinant of information

asymmetry, and of the

profitability of insider

trades.



Figure 23: Excess Returns of Insider Purchases Split by R&D Intensity


Our results, depicted in Figure 23, confirm academic findings. In our sample, we find a substantial difference in the performance of insider trades depending on the R&D Intensity, both in the short- and the long-term. Indeed, purchases from insiders in highly R&D intensive companies generate about 10% return more than the benchmark, where the number drops to less than 3% for less R&D intensive firms. This corroborates the argument that insider trades really act as an “instrument” with which private information is disseminated to the market.

Earnings Surprises While an earnings announcement enlightens the market regarding recent company performance, the way this announcement is interpreted is not in the hands of the management. Consequently, additional actions taken by top executives, such as insider trades, could serve either as a confirmatory signal, or, on the contrary, as an “adjustment” signal to the information conveyed in the earnings statement. Allen and Ramanan (1995) found that insider purchases acted as confirmatory signals for positive earnings surprises, similarly to Giamouridis et al. (2008). In what follows, we will define the Earnings Surprise as the excess return of the stock relative to its benchmark from the day before the earnings announcement to four days after the announcement, in line with Giamouridis et al. (2008). As we outlined in our report on earnings surprises (see Mesomeris et al. (2011)), the advantage of this metric is to capture information over and above what is contained in the raw EPS number: it could be tangible or intangible information, regarding management changes, margins, future prospects, etc. It also means we are capturing the market interpretation of the earnings announcement.

Purchases from insiders in

highly R&D intensive

companies generate about

10% return more than the

benchmark, where the

number drops to less than

3% for less R&D intensive

firms.

Actions taken by top

executives following

earnings surprises, such as

insider trades, could serve

either as a confirmatory

signal, or as an

“adjustment” signal to the

information conveyed in the

earnings statement.



Figure 24: Excess Return of Insider Purchases Split by Earnings Surprise Threshold


In Figure 24, we show the results of an event study where trades are divided into four categories: an excess return around the announcement day below -5%, between -5% and 0%, between 0% and 5%, and above the 5% threshold. It is very interesting to note that, while we seem to be confirming the findings of previous research (namely that insider purchases after positive earnings surprises prove informative), we observe additionally that insider purchases following negative earnings surprises are also generating positive excess returns. This second point suggests that insiders seem to act as contrarian traders, “buying the dip” in their own company following some underperformance post a negative earnings surprise. This behavior has actually been observed by Lakonishok and Lee (2001) on an aggregate basis. We would also point out that this finding suggests a non-linear relationship between insider trades and previous earnings surprises, a feature not captured in the previous research papers we have mentioned from a methodological point of view.

Short Interest Measures of market sentiment that would be quite interesting to analyze in the context of insider trading activity also emanate from the stock lending market. As we have shown in past research (see Cahan et al. (2011) and Avettand-Fenoel et al. (2012)), short sellers are usually more informed than typical equity market participants, and thus their activity can be useful in conveying the overall market feeling surrounding a stock. Similarly to what we have done with Earnings Surprises, we investigate whether insider trading constitutes a confirmatory signal in the presence of short interest, as has been found in Purnanandam and Seyhun (2011). In what follows, we will use, in line with Purnanandam and Seyhun (2011), the innovation in short interest for a given stock (defined as the z-score over the year of the ratio of the number of shares on loan divided by the number of shares outstanding) as well as the short interest level itself.

While we seem to be

confirming the findings of

previous research (namely

that insider purchases

confirming positive earnings

surprises prove informative),

we observe additionally that

insider purchases following

negative earnings surprises

are also generating positive

abnormal returns.

We investigate here whether

insider trading constitutes a

confirmatory signal to the

one conveyed by short

sellers activity.




Short Interest Level


Short Interest Innovation


First of all, it is interesting to notice in Figure 26 that our results are in line with Pumanandam and Seyhun (2011). That particular event study suggests that, when short sellers increase their position (i.e. take on a more negative view), insider purchases are not profitable, and vice versa. In other words, a decrease in the short selling activity for a given stock is usually a positive sign, and an insider purchasing the stock acts as a confirmatory signal, leading to positive abnormal returns. Nevertheless, one would also notice that the picture is somewhat different in Figure 25: similarly to the contrarian behavior we observed previously with earnings surprises, it seems that a high level of short selling activity entices insiders to buy their own stock to counteract the prevailing market sentiment, which leads to positive excess share price returns. Transactions on stocks with a low level of short interest don’t prove informative in the long run.

Conclusion We now summarize several key points drawing from the event studies we have conducted. It appears quite clearly that different insider trades convey information to the market with varying success. For each of the metrics analyzed above, we have effectively identified a subset of trades which generated significantly higher abnormal returns. This identification process actually relies mainly on four ideas:

Limited Attention: In not less than three different metrics, we have learnt that not opting for the obvious trade was rewarding. It would seem rather logical that most of the investors’ attention is devoted to trades by Chairmen or CEOs, and trades that are large in size, but these trades actually perform rather poorly, potentially on the back of a high price impact and quick arbitrage. In a similar spirit, isolated trades (i.e. when an insider does not trade several months in a row) are characterized by underreaction, while they actually usually convey more positive information than sequenced trades.

Insider Dynamics: Knowing that an insider is buying a stock is one thing, but knowing what she, or her colleagues, has done in the recent past is another. We found measures of the evolution of insider trading activity over a number of months like the NIVR and NICR to be useful in understanding whether the insider has been acting in line with other executives (in which case the trade conveys a sense of certainty through “herding” and on average leads to superior returns), or she has been taking a different stance. In addition, whether a trade is isolated or part of a sequence of transactions also helps us understand if the insider is in some way behaving like a “noise trader” (thus not conveying additional information with more trades).

When looking at the level of

short interest, insider trades

prove contrarian. However,

when the innovation in short

interest is examined, insider

trades seem to confirm the

change in sentiment.

Metrics used to classify

insider trades can be split

into four categories: Limited

Attention, Insider Dynamics,

Information Asymmetry, and

Confirmatory/Contrarian

Trading.



Information Asymmetry: Insiders may be acting upon information non-disclosed by the firm to the public, which is termed as information asymmetry. We found that the more asymmetric the information likely is between the insider/company and the market, the higher the information content of the insider purchase. Therefore, to the extent that the insider’s position in the pecking order can be viewed as a proxy for the degree of information asymmetry, we observe that information hierarchy is reflected in future returns (except for the most “followed” directors). Secondly, R&D intensity is a very good proxy for information asymmetry, and clearly impacts the future performance of insider purchases.

Confirmatory/Contrarian Trading: Insider purchases can also take place as a reaction to the prevailing market sentiment towards the particular company. The relationship between the two has been investigated in Hagenau et al. (2012), where they found insiders to be buying (selling) following bad (good) corporate as well as third-party news, i.e. a contrarian behavior. In our event studies above, we highlighted both contrarian and confirmatory behavior. Indeed, insiders are buying and generating positive returns after negative earnings surprises and when the firm’s stocks are heavily shorted (i.e. acting as contrarian investors). Nevertheless, we also found trades to be meaningful after positive earnings surprises and when short sellers were decreasing their positions (i.e. confirmatory).

On a separate note, we would like to point out that all the event studies and analysis above have focused purely on the selection of insiders’ purchases, and not of insiders’ sales. The reason for not dwelling too much on sales is that that too few metrics are useful in segregating underperforming from non-performing insider sales. In a number of cases, we find the categories used above to lead to overlapping confidence intervals, suggesting that the metrics are not adequate to statistically distinguish between sell trades. Only in the case of Market Cap and Short Interest Level were we able to find a significant difference, but the results were not economically significant and thus not useful in the context of a trading strategy. We thus continue focusing on insider purchases in the next sections.



Assessing Selection Criteria via Classification

In the previous section, we investigated the selection of insider trades in a rather qualitative manner: our choice of categories in the event studies was theoretically motivated, perhaps with the benefit of some hindsight. Furthermore, all these studies were computed in a univariate manner, meaning that only the metric involved in the event study was used for the selection of potentially promising signals. One obvious question would be whether the results hold once the threshold used to split the data by each metric is not pre-determined, and also whether the various metrics can be used in conjunction.

We use here a technique called FOREST4 to answer both questions. FOREST is a machine learning algorithm which aims at classifying observations into various categories using a multitude of decision trees. The algorithm is fed with observations on a number of variables (in our case, these will be the insider level, trade size, etc. of all insider purchases) and a dependent variable that we try to explain and/or predict, the forward excess returns. To make the learning robust, it is often recommended to use classification rather than regression, i.e. instead of trying to predict the level of excess returns, one might attempt to predict whether the excess returns are “very bad”, “bad”, “neutral”, “good”, or “very good”.

As explanatory variables, we utilize the following metrics: Director Level (as the A/B/C categories explained previously), Size (using deciles for the transaction value), NIVR (as defined above), IsolSeq (the number of consecutive months during which the insider traded), R&D Intensity, Earnings Surprises, and the level of Short Interest. The FOREST algorithm is trained using the 12-month forward excess returns bucketed in 5 quintiles. Initially, our sample consists of 10,000 buy trades selected randomly from our dataset, and we assess whether any overfitting has occurred using the remaining data. Note that due to data availability issues on R&D Intensity and Short Interest (mainly), the total sample size consists of 13,566 trades (i.e. 3,566 trades are left for out-of-sample evaluation).

Figure 27: “Confusion Tables” for Actual vs. Predicted Classes and Classification Errors In-sample (10,000 trades) Out-of-sample (3,566 trades)

Predicted class Predicted class

1 2 3 4 5 Error 1 2 3 4 5 Error

Actual class

1 914 127 95 85 119 31.8% Actual class

1 345 37 14 19 14 19.6%

2 128 900 269 81 111 39.6% 2 51 315 60 31 14 33.1%

3 53 188 1142 222 123 33.9% 3 40 96 405 116 41 42.0%

4 47 80 287 816 272 45.7% 4 17 38 75 298 55 38.3%

5 55 66 112 169 1346 23.0% 5 56 43 50 100 478 34.3%Source: 2iQ Research, Factset, Deutsche Bank

In Figure 27, we observe that the relationships uncovered previously in the event studies were not spurious at all, and are confirmed when using the FOREST algorithm. Using our seven metrics, the algorithm is able to adequately distinguish between convincing trades (class 5) and unconvincing ones (class 1) with a fairly manageable degree of error, both in and out of sample (a classification error of 33.3% means that 1 out of 3 trades is misclassified).

Knowing that our classification algorithm detects trades correctly is one thing, but the way it does it is another. Hopefully the variations induced by a given variable on the forward returns will follow closely our qualitative decision process above. To analyze this particular point, we use partial dependence plots. A partial dependence plot is a visual description of the marginal effect of a variable on the probability of a trade being classified in a given class. We will focus here on class 5 (trades with the best forward excess returns).

4 More details on the FOREST algorithm can be found in Breiman (2001).

One obvious question is

whether the results hold

once the threshold used to

split the data by each metric

is not pre-determined. Also,

can we combine the various

metrics together?

We use the FOREST

algorithm to classify trades

and try and explain their 12-

month forward returns,

using seven explanatory

variables.

We train the FOREST using

10,000 trades selected

randomly, while the out-of-

sample set consists of 3,566

trades.

The “confusion tables”

show relatively small

prediction errors (about 1

out of 3 high conviction

trade is misclassified).

We use partial dependence

plots to analyze the

variations induced by a

given variable on the

forward returns.



Figure 28: Partial Dependence Plot for Level Figure 29: Partial Dependence Plot for Transaction Size


Figure 30: Partial Dependence Plot for NIVR Figure 31: Partial Dependence Plot for IsolSeq


Figure 32: Partial Dependence Plot for R&D Intensity Figure 33: Partial Dependence Plot for Earnings Surprise




Figure 34: Partial Dependence Plot for Short Interest

Level

Figure 35: Variable Importance in the FOREST


We show from Figure 28 to Figure 34 the partial dependence plots for our seven explanatory variables. Without going into too much detail, each of them confirm our previous findings to some extent, if not entirely. First of all, the Level factor confirms that a Level B insider trade will lead to higher future returns. Interestingly, for Transaction Size, NIVR and IsolSeq, the relationship between the metric and the marginal dependence is nicely monotonic: smaller trades lead to higher returns, and higher NIVR trades (in particular for NIVR = 1) do so too. Regarding IsolSeq, it confirms isolated trades perform the best. The dependence on our last three variables appears noisier: it is thus preferable to look at the smoothed curves to extrapolate the general trend. On the one hand, we confirm again that higher R&D Intensity and a higher Short Interest level lead to higher returns. On the other hand, the non-linearity feature with regards to Earnings Surprises we highlighted previously (i.e. trades perform better when the Earnings Surprise is outside the [-5%; 5%] bound) is nicely captured by the FOREST algorithm.

To conclude on the use of FOREST, we plot in Figure 35 the relative importance of the seven variables in our model, as computed by the mean decrease in the Gini index. This shows us that R&D Intensity, Earnings Surprises, and Short Interest are the three most important variables for the classification problem at hand. Yet, the four other variables are not far behind, suggesting that all variables are useful.

The partial dependence

plots of our seven

explanatory variables

confirm our previous

findings to some extent, if

not totally.

In particular, the non-linear

effect on Earnings Surprise

is nicely captured.

R&D Intensity, Earnings

Surprises, and Short Interest

are the three most

important variables for our

classification problem.



Outperforming with Insider Information Satellite Insider Portfolios

As we have mentioned in the beginning of the report, implementing traditional quant signals based on insider trading is not an easy task. Indeed, strategies using directors’ dealings should rather be classified as event driven strategies, since not all companies have insiders operating trades. This means that in a universe of 1000 stocks, only a certain number of them would produce signals, which makes it hard to create diversified baskets exploiting premiums cross-sectionally.

We use a simple methodology where we leverage the information contained in insider trades by lining up a satellite portfolio alongside the benchmark index, with the intention of outperforming the benchmark. Such a methodology has as the advantage that the weight allocated to the insider strategy can be tuned to target a given tracking error (see Figure 36, where α is the weight allocated to the Insider strategy). Based on the unconditional backtest analyzed in the second section of this report, we will use, for now, a 25% allocation to the insider strategy, allocating the remaining 75% to the benchmark index. This leads to relatively small tracking errors of 2% to 3% per annum. The insider strategy encompasses the selected high conviction purchases using the metrics analyzed in detail in the previous Section.

Figure 36: Schematic View of the Portfolio Construction Process

All Insider Purchases

S&P BMI EuropeBenchmark

α

1-α

Final Portfolio

Selected InsiderPurchases

Source: Deutsche Bank

We carry out a number of backtests where trades are selected on the basis of seven signals, first independently, and then in conjunction. Trades are first scored from -1 to 1 on each signal. If data availability allows multiple signals to be considered, scores are averaged across the signals (with equal weights). Finally, the top N stocks are selected to construct the equally-weighted insider portfolio. Note than in the case where more than N stocks have the same (highest) score, all stocks with the highest score are selected. We decide to set N equal to 20: a lower number would increase the impact of estimation error, while a higher number would make portfolios probably too large when considering that longer holding periods lead to more overlapping portfolios. For instance, with a 6-month holding period, we would have 6 overlapping portfolios and thus 120 stocks in our insider portfolio at each point in time. Robustness tests show that varying N doesn’t impact the performance of the strategy too much. We compare stock portfolios whereby the selection criteria are utilized with a portfolio (“No Selection”) of all insider purchase stocks.

We use a simple

methodology to leverage off

the information contained in

insider trades, lining up a

satellite portfolio alongside

the benchmark index, with

the intention of

outperforming the

benchmark.

Before fine-tuning the

weights allocated to the

insider strategy, we initially

use a 25% allocation.

We run a number of

backtests where we select

20 trades on the basis of

seven signals, first

independently, and then in

conjunction.



Figure 37: Information Ratio for Selective Insider Trade Strategies, allocated at 25%, for 1, 3, 6, and 12-month holding

periods


From Figure 37, we understand that at the 1-month horizon, selecting trades is better in some but not in all cases, since three factors are doing worse than the No Selection portfolio. As we extend the holding period, being selective seems to add value as for 6- and 12-month holding periods, all the selection metrics seem to work better than the No Selection portfolio, and even more so as the IR of the No Selection portfolio decreases.

In the next step, we decide to combine all the scores together to create a unique ranking of insider purchases. Considering that our trade selection process has been focused more on longer term than 1-month ahead returns, in what follows we will only use a 6-month holding period for the sake of brevity. Furthermore, instead of sticking to a 25% allocation for the insiders trade strategies, we prefer here to carefully select the allocation in order to target a given tracking error. This can easily be done using the following formula:

)()'( BIBI wwwwTE

−Σ−=α

Where TE is our target tracking error, Iw and Bw the stock weights in the Insiders and Benchmark portfolios respectively, Σ the variance-covariance matrix. We decide here to choose a target tracking error of 2%. Empirically, this leads to an allocation of between 15% to 46% of the final portfolio weight.

For the 6 and 12-month

holding periods, all the

selection processes work

better than the original

portfolio.

Instead of sticking to a 25%

allocation for the insider

trade strategies, we prefer

here to carefully select the

allocation in order to target

a given tracking error.



Figure 38: Strategy Performance without Factors (No

Selection) and with All Factors

Figure 39: Cumulative Excess Return of the Strategies

without Factors (No Selection) and with All Factors

No Selection All Factors

Before cost

Excess Returns 1.01% 1.81%

Tracking Error 2.23% 2.16%

Information Ratio 0.45 0.84

After cost

Excess Returns 0.80% 1.58%

Tracking Error 2.23% 2.16%

Information Ratio 0.36 0.73

Transaction costs are assumed to be 20bps each way. Source: 2iQ Research, Factset, Deutsche Bank Source: 2iQ Research, Factset, Deutsche Bank

The results in Figure 38 and Figure 39 confirm the effectiveness of the trade selection process. Indeed, before costs, we are able to achieve 80% more return for a slightly smaller tracking error, thus leading to almost double the information ratio. After costs, which we assume to be of 20 bps each way, results are still positive for the trades selected on all factors, with an information ratio slightly more than doubled. When comparing the cumulative excess return of the two strategies, we observe that both experienced a drawdown from mid-2007 to early 2009 (potentially reflecting a Value exposure at that time). However, while the No Selection strategy was almost flat from mid-2009 to the recent past, our portfolio conditioned on All Factors continued outperforming over the last 3 years.

Robustness of the Insider Strategy Returns

Before concluding this report, it is important to determine a couple of more things: first of all, whether the selection criteria we have utilized end up producing returns which are the consequence of some random behavior, and secondly, whether we are merely capturing typical risk factor exposure such as Size, Value or Momentum.

On the former question, it is relatively easy to understand if our insider portfolio with a “smart” selection process is really different from a portfolio of purchases selected at random. For this purpose, we conduct 1000 backtests, where the top 20 stocks are selected at random (we thus assign a random score to each trade to make the selection).

Before costs, we are able to

achieve 80% more return for

a slightly smaller tracking

error, while after costs, the

information ratio comes out

as slightly more than

doubled.

We want to assess whether

the selection criteria we

have utilized really add

value, or whether they

produce returns which

cannot be differentiated

from random behavior, and

secondly whether we are

capturing typical risk factor

exposures.



Figure 40: IR Distribution of the Randomly Selected Trades Portfolios and IRs with and

without Selection of Insider Trades on the Factors


Figure 40 shows clearly that had the trades been selected at random, the Information Ratio of our final portfolio would have been closer to 0.6 rather than falling above the 99th percentile of the distribution.

On the latter question, our methodology for checking our exposures to the market (MKT), Size (SMB), Value (HML) and Momentum (MOM) factors is standard: we first build the four factors in our S&P BMI Europe universe, and then carry out a regression analysis with the insider strategy returns as the dependent variable.

Figure 41: Fama-French and Carhart Regression on Insiders Strategies Intercept MKT SMB HML MOM

No Selection Estimate 0.077% 0.047 0.131 -0.022

t-stat 1.84 4.64*** 7.00*** -1.45

Estimate 0.077% 0.048 0.128 -0.020 0.06

t-stat 1.85 4.68*** 6.84*** -1.24 1.25

All Factors Estimate 0.137% 0.046 0.094 -0.010

t-stat 3.00** 4.15*** 4.57*** -0.60

Estimate 0.137% 0.047 0.091 -0.008 0.05

t-stat 3.00** 4.17*** 4.44*** -0.44 0.92Source: 2iQ Research, Factset, Deutsche Bank

Figure 41 shows that apart from a significant exposure to the SMB factor, there is no relationship with the HML and MOM factors; the MKT beta is significant but relatively small (less than 0.05). The SMB exposure actually has a simple explanation: our Selected Insider Trade portfolio is composed of 20 equally-weighted stocks, overweighting by construction small versus large stocks. A potential solution to this drawback would be to weigh stocks by their market capitalization in the Selected Insider Trades portfolio. Finally, the estimated alphas in Figure 41 are not statistically significant when we don’t select trades, but statistically significant at the 99% significance level when we select trades using all factors investigated.

If we generate 1000

portfolios where 20 trades

are selected at random,

most of these perform much

more poorly, suggesting a

very low probability of our

selection process being

random.

We check our exposure to

MKT, SMB, HML and MOM

factors by estimating Fama-

French and Carhart model

regressions.

Apart from SMB, no

particular exposure can be

observed. SMB can be

explained by the equally-

weighted scheme.

Estimated alphas for the

“No Selection” insider

purchases portfolio are

statistically insignificant,

while alphas of the portfolio

whereby insider trades are

selected on the basis of all

factors are statistically

significant.



References Aboody, D. and Lev, B., 2000, "Information asymmetry, R&D, and insider gains", The Journal of Finance, Vol. 55, No. 6.

Allen, S. and Ramanan, R., 1995, "Insider trading, earnings changes, and stock prices", Management Science, Vol. 41, No. 4.

Avettand-Fenoel, J.-R. et al., 2012, "Thematic Report - Introducing the 3S Model", Deutsche Bank Equity Quantitative Strategy, 26th March 2012.

Barclay, M.J. and Warner, J.B., 1993, "Stealth trading and volatility: Which trades move prices?", Journal of Financial Economics, Vol. 34, No. 3.

Breiman, L., 2001, "Random Forests", Machine Learning, Vol. 45, No. 1.

Cahan, R. et al., 2011, "Signal Processing - The long and the short of it", Deutsche Bank Equity Quantitative Strategy, 18th January 2011.

Cicero, D. and Wintoki, M., 2012, "Insider Trading Patterns", SSRN Working Papers.

Dardas, K., 2011, "Identifying Profitable Insider Transactions", SSRN Working Papers.

Dardas, K. and Guttler, A., 2011, "Are Directors' Dealings Informative? Evidence from European Stock Markets", Financial Markets and Portfolio Management, Vol. 25.

Giamouridis, D., Liodaki, M. and Moniz, A., 2008, "Some insiders are indeed smart investors", SSRN Working Papers.

Hagenau, M., Korczak, A. and Neumann, D., 2012, "Buy on Bad News, Sell on Good News: How Insider Trading Analysis Can Benefit from Textual Analysis of Corporate Disclosures", SSRN Working Papers.

Jeng, L. A., Metrick, A. and Zeckhauser, R., 1999, "The profits to insider trading: A performance-evaluation perspective", NBER Working Paper No. 6913.

Kyle, A., 1985, "Continuous Auctions and Insider Trading", Econometrica, Vol. 53, No. 6.

Lakonishok, J. and Lee, I., 2001, "Are insider trades informative?", Review of Financial Studies, Vol. 14, No. 1.

Mesomeris, S. et al., 2011, "Quantitative Musing - Earnings Surprises: It's All About Beat or Miss", Deutsche Bank Equity Quantitative Strategy, 8th June 2012.

Purnanandam, A. and Seyhun, N. H., 2011, "Do short sellers trade on private information or false information?", SSRN Working Papers.

Salvini, M. et al., 2012, "Quantitative Musing - Share Buybacks: Is it Worth it?", Deutsche Bank Equity Quantitative Strategy, 14th March 2012.

Seyhun, H. N., 1986, "Insiders' Profits, Costs of Trading, and Market Efficiency", Journal of Financial Economics, Vol. 16, No. 2.



Appendix 1 Important Disclosures Additional information available upon request

For disclosures pertaining to recommendations or estimates made on a security mentioned in this report, please see the most recently published company report or visit our global disclosure look-up page on our website at http://gm.db.com/ger/disclosure/DisclosureDirectory.eqsr.

Analyst Certification The views expressed in this report accurately reflect the personal views of the undersigned lead analyst(s). In addition, the undersigned lead analyst(s) has not and will not receive any compensation for providing a specific recommendation or view in this report. Jean-Robert Avettand-Fenoel/Spyros Mesomeris/Marco Salvini/Yiyi Wang

Hypothetical Disclaimer Backtested, hypothetical or simulated performance results have inherent limitations. Unlike an actual performance record based on trading actual client portfolios, simulated results are achieved by means of the retroactive application of a backtested model itself designed with the benefit of hindsight. Taking into account historical events the backtesting of performance also differs from actual account performance because an actual investment strategy may be adjusted any time, for any reason, including a response to material, economic or market factors. The backtested performance includes hypothetical results that do not reflect the reinvestment of dividends and other earnings or the deduction of advisory fees, brokerage or other commissions, and any other expenses that a client would have paid or actually paid. No representation is made that any trading strategy or account will or is likely to achieve profits or losses similar to those shown. Alternative modeling techniques or assumptions might produce significantly different results and prove to be more appropriate. Past hypothetical backtest results are neither an indicator nor guarantee of future returns. Actual results will vary, perhaps materially, from the analysis.



Regulatory Disclosures

1. Important Additional Conflict Disclosures

Aside from within this report, important conflict disclosures can also be found at https://gm.db.com/equities under the "Disclosures Lookup" and "Legal" tabs. Investors are strongly encouraged to review this information before investing.

2. Short-Term Trade Ideas

Deutsche Bank equity research analysts sometimes have shorter-term trade ideas (known as SOLAR ideas) that are consistent or inconsistent with Deutsche Bank's existing longer term ratings. These trade ideas can be found at the SOLAR link at http://gm.db.com.

3. Country-Specific Disclosures

Australia & New Zealand: This research, and any access to it, is intended only for "wholesale clients" within the meaning of the Australian Corporations Act and New Zealand Financial Advisors Act respectively. Brazil: The views expressed above accurately reflect personal views of the authors about the subject company(ies) and its(their) securities, including in relation to Deutsche Bank. The compensation of the equity research analyst(s) is indirectly affected by revenues deriving from the business and financial transactions of Deutsche Bank. In cases where at least one Brazil based analyst (identified by a phone number starting with +55 country code) has taken part in the preparation of this research report, the Brazil based analyst whose name appears first assumes primary responsibility for its content from a Brazilian regulatory perspective and for its compliance with CVM Instruction # 483. EU countries: Disclosures relating to our obligations under MiFiD can be found at http://www.globalmarkets.db.com/riskdisclosures. Japan: Disclosures under the Financial Instruments and Exchange Law: Company name - Deutsche Securities Inc. Registration number - Registered as a financial instruments dealer by the Head of the Kanto Local Finance Bureau (Kinsho) No. 117. Member of associations: JSDA, Type II Financial Instruments Firms Association, The Financial Futures Association of Japan, Japan Investment Advisers Association. Commissions and risks involved in stock transactions - for stock transactions, we charge stock commissions and consumption tax by multiplying the transaction amount by the commission rate agreed with each customer. Stock transactions can lead to losses as a result of share price fluctuations and other factors. Transactions in foreign stocks can lead to additional losses stemming from foreign exchange fluctuations. "Moody's", "Standard & Poor's", and "Fitch" mentioned in this report are not registered credit rating agencies in Japan unless “Japan” or "Nippon" is specifically designated in the name of the entity. Russia: This information, interpretation and opinions submitted herein are not in the context of, and do not constitute, any appraisal or evaluation activity requiring a license in the Russian Federation.


European location

Deutsche Bank AG London 1 Great Winchester Street London EC2N 2EQ Tel: (44) 20 7545 8000

Deutsche-Bank AG, 3, Avenue de Friedland 75008 Paris Cedex 8 France Tel: (33) 1 44 95 64 00

Deutsche Bank AG Equity Research Große Gallusstraße 10-14 60272 Frankfurt am Main Germany Tel: (49) 69 910 00

Deutsche Bank Sim S.p.a Via Santa Margherita 4 20123 Milan Italy Tel: (39) 0 24 024 1

Deutsche Bank AG Herengracht 450 1017 CA Amsterdam Netherlands Tel: (31) 20 555 4911

Deutsche Securities S.V.B, S.A. Paseo de la Castellana, 18 Entreplanta 28046 Madrid, Spain Tel: (34) 91 335 5900

Deutsche Bank AG Stureplan 4 A, Box 5781 S-114 87 Stockholm Sweden Tel: (46) 8 463 5500

Deutsche Bank AG Uraniastrasse 9 PO Box 7370 8023 Zürich Switzerland Tel: (41) 1 224 5000

Deutsche Bank AG, Helsinki Kaivokatu 10 A, P.O.Box 650 FI-00101 Helsinki Finland Tel: (358) 9 25 25 25 0

Deutsche Bank AG Hohenstaufengasse 4 1010 Vienna Austria Tel: (43) 1 5318 10

Deutsche Bank Ltd. Aurora business park 82 bld.2 Sadovnicheskaya street Moscow, 115035 Russia Tel: (7) 495 797-5000

Deutsche Bank AG, Warsaw al.Armii Ludowej 26 Budynek FOCUS 00-609 Warsaw Poland Tel: (48) 22 579 87 00

Deutsche Bank AG, Turkey Eski Buyukdere Cad. Tekfen Tower No:209 Kat:17-18 TR-34394 Istanbul Tel: (90) 212 317 01 00

Deutsche Bank AG, Greece 23A Vassilissis Sofias Avenue 6th Floor 10674 Athens, Greece Tel: (30) 210 72 56 150

International Locations

Deutsche Bank Securities Inc. 60 Wall Street New York, NY 10005 United States of America Tel: (1) 212 250 2500

Deutsche Bank AG London 1 Great Winchester Street London EC2N 2EQ United Kingdom Tel: (44) 20 7545 8000

Deutsche Bank AG Große Gallusstraße 10-14 60272 Frankfurt am Main Germany Tel: (49) 69 910 00

Deutsche Bank AG Deutsche Bank Place Level 16 Corner of Hunter & Phillip Streets Sydney, NSW 2000 Australia Tel: (61) 2 8258 1234

Deutsche Bank AG Filiale Hongkong International Commerce Centre, 1 Austin Road West,Kowloon, Hong Kong Tel: (852) 2203 8888

Deutsche Securities Inc. 2-11-1 Nagatacho Sanno Park Tower Chiyoda-ku, Tokyo 100-6171 Japan Tel: (81) 3 5156 6770

GRCM2012PROD026992

Disclaimer

The information and opinions in this report were prepared by Deutsche Bank AG or one of its affiliates (collectively "Deutsche Bank"). The information herein is believed to be reliable and has been obtained from public sources believed to be reliable. Deutsche Bank makes no representation as to the accuracy or completeness of such information. Deutsche Bank may engage in securities transactions, on a proprietary basis or otherwise, in a manner inconsistent with the view taken in this research report. In addition, others within Deutsche Bank, including strategists and sales staff, may take a view that is inconsistent with that taken in this research report. Deutsche Bank may be an issuer, advisor, manager, distributor or administrator of, or provide other services to, an ETF included in this report, for which it receives compensation. Opinions, estimates and projections in this report constitute the current judgement of the author as of the date of this report. They do not necessarily reflect the opinions of Deutsche Bank and are subject to change without notice. Deutsche Bank has no obligation to update, modify or amend this report or to otherwise notify a recipient thereof in the event that any opinion, forecast or estimate set forth herein, changes or subsequently becomes inaccurate. Prices and availability of financial instruments are subject to change without notice. This report is provided for informational purposes only. It is not an offer or a solicitation of an offer to buy or sell any financial instruments or to participate in any particular trading strategy. Target prices are inherently imprecise and a product of the analyst judgement. As a result of Deutsche Bank’s March 2010 acquisition of BHF-Bank AG, a security may be covered by more than one analyst within the Deutsche Bank group. Each of these analysts may use differing methodologies to value the security; as a result, the recommendations may differ and the price targets and estimates of each may vary widely. In August 2009, Deutsche Bank instituted a new policy whereby analysts may choose not to set or maintain a target price of certain issuers under coverage with a Hold rating. In particular, this will typically occur for "Hold" rated stocks having a market cap smaller than most other companies in its sector or region. We believe that such policy will allow us to make best use of our resources. Please visit our website at http://gm.db.com to determine the target price of any stock. The financial instruments discussed in this report may not be suitable for all investors and investors must make their own informed investment decisions. Stock transactions can lead to losses as a result of price fluctuations and other factors. If a financial instrument is denominated in a currency other than an investor's currency, a change in exchange rates may adversely affect the investment. All prices are those current at the end of the previous trading session unless otherwise indicated. Prices are sourced from local exchanges via Reuters, Bloomberg and other vendors. Data is sourced from Deutsche Bank and subject companies. Past performance is not necessarily indicative of future results. Deutsche Bank may with respect to securities covered by this report, sell to or buy from customers on a principal basis, and consider this report in deciding to trade on a proprietary basis. Derivative transactions involve numerous risks including, among others, market, counterparty default and illiquidity risk. The appropriateness or otherwise of these products for use by investors is dependent on the investors' own circumstances including their tax position, their regulatory environment and the nature of their other assets and liabilities and as such investors should take expert legal and financial advice before entering into any transaction similar to or inspired by the contents of this publication. Trading in options involves risk and is not suitable for all investors. Prior to buying or selling an option investors must review the "Characteristics and Risks of Standardized Options," at http://www.theocc.com/components/docs/riskstoc.pdf If you are unable to access the website please contact Deutsche Bank AG at +1 (212) 250-7994, for a copy of this important document. The risk of loss in futures trading, foreign or domestic, can be substantial. As a result of the high degree of leverage obtainable in futures trading, losses may be incurred that are greater than the amount of funds initially deposited. Unless governing law provides otherwise, all transactions should be executed through the Deutsche Bank entity in the investor's home jurisdiction. In the U.S. this report is approved and/or distributed by Deutsche Bank Securities Inc., a member of the NYSE, the NASD, NFA and SIPC. In Germany this report is approved and/or communicated by Deutsche Bank AG Frankfurt authorized by the BaFin. In the United Kingdom this report is approved and/or communicated by Deutsche Bank AG London, a member of the London Stock Exchange and regulated by the Financial Services Authority for the conduct of investment business in the UK and authorized by the BaFin. This report is distributed in Hong Kong by Deutsche Bank AG, Hong Kong Branch, in Korea by Deutsche Securities Korea Co. This report is distributed in Singapore by Deutsche Bank AG, Singapore Branch, and recipients in Singapore of this report are to contact Deutsche Bank AG, Singapore Branch in respect of any matters arising from, or in connection with, this report. Where this report is issued or promulgated in Singapore to a person who is not an accredited investor, expert investor or institutional investor (as defined in the applicable Singapore laws and regulations), Deutsche Bank AG, Singapore Branch accepts legal responsibility to such person for the contents of this report. In Japan this report is approved and/or distributed by Deutsche Securities Inc. The information contained in this report does not constitute the provision of investment advice. In Australia, retail clients should obtain a copy of a Product Disclosure Statement (PDS) relating to any financial product referred to in this report and consider the PDS before making any decision about whether to acquire the product. Deutsche Bank AG Johannesburg is incorporated in the Federal Republic of Germany (Branch Register Number in South Africa: 1998/003298/10). Additional information relative to securities, other financial products or issuers discussed in this report is available upon request. This report may not be reproduced, distributed or published by any person for any purpose without Deutsche Bank's prior written consent. Please cite source when quoting. Copyright © 2012 Deutsche Bank AG

DB European Quant Strategy - QM - Are Insiders Alpha Generators 20120926

Documents

Transcript of DB European Quant Strategy - QM - Are Insiders Alpha Generators 20120926