The Immediate Impact of SEC Filings on Volume and Volatility David Grossman 1 1 David Grossman is a member of the Class of 2017 at the University of Chicago. UChicago Undergraduate Business Journal Abstract In this paper, I use a stratified single variable regression analysis to explore the causal impact of the publishing of financial filings with the Securities and Exchange Commission (SEC) on real time market prices and volume. The intended purpose of SEC filing requirements is to encourage transparent dissemination of relevant information to shareholders. My work aims to evaluate whether entities in the market actively monitor the SEC in order to be the first to trade on recent filing information. I found that the average filing is associated with a 0.0091% change in stock price in the first five minutes after its posting but that this relationship is not statistically significant. More importantly, I found a highly statistically significant relationship (p<0.001) between filings and volume. On average, I found that a filing results in a 36.7% increase in volume traded in the first five minutes after its posting. My findings indicate that markets are more responsive to filings which contain new information about the company than they are to notifications of trades made by other investors. In addition, markets are more sensitive to disclosure of trading by company insiders than to disclosure of large acquisitions by institutions. Lastly, I tested for the existence of insider trading and information leakage: the market reaction to SEC filings beginning just prior to their actual publication. While I did find weak cumulative evidence of insider trading based on the direction and magnitude of the observed effect, that effect was not statistically significant. Spring 2015 1 UChicago Undergraduate Business Journal Introduction and Literature Review There has been a plethora of research done on what moves markets. This includes research focused on the SEC. Specifically, there have been studies conducted on the impact of SEC-mandated market risk disclosures, market reactions to SEC filings correcting previous statements, and the benefits of SEC accounting standards requirements (Feroz et al., 2008; Roulstone, 1999; Linsmeier, 2002). Feroz et al. found that outcomes of SEC investigations into accounting fraud have significant impact on market prices (Feroz et. al 2008). While it is possible that traders only pay attention to SEC filings when it turns out that they have been falsified, these findings suggest that traders are likely make financial bets based on SEC filings, the thesis of which no longer holds when the numbers are revised – resulting in a reversal of that bet and an associated market fluctuation. Because this requires traders to pay attention to SEC filings in the first place, there is reason to examine the realtime impact of SEC filings when they first become available in the marketplace. An observed decrease in longterm volatility following detailed discussion of risk in published 10-K (comprehensive annual report) filings provides additional evidence that markets pay attention to SEC filings (Linsmeier, 2002). However, no one has yet researched whether SEC filings have real-time implications for stock volume and volatility (prices) because the data was not available. The SEC does not provide a timestamp for when a filing is made public, only for when it is electronically filed with the SEC, so previous researchers did not have access to the actual timestamp of when the information contained in SEC filings became available to the markets. Therefore, to facilitate this research, I coded a scanner of the SEC website which constantly monitored the Edgar database for updates, and collected the data myself. Due to the infrequency of tangible new market information, I am confident that given a short enough time window around the publishing of a filing, any non-random relative price movements or increases in volume traded could be directly attributed to the filing. Conservatively, I picked an observation time frame of ten minutes, comparing the five minutes of stock market data after a filing’s posting to the five minutes prior. Lastly, I attempted to determine if there was any insider trading happening through the SEC. Given the value of having early access to market information, there is an incentive for someone at the SEC to distribute the filing to interested parties before posting it publicly. If a bribe is acted upon then there could be a non-random relationship between how long it takes the SEC to post a filing after receiving it and the observed impact of the filing. Presumably, the longer it takes to post the greater the chance information will be leaked early and the more time insider traders have to act on that information. The more the information contained in the filing is priced into the market already at the time of ‘official’ posting, the less of an observed impact it would have. So in the case of insider trading, I would expect a small (insider traders constitute a small fraction of the market) negative relationship between how long it took to post a filing after receiving it and the magnitude of volume fluctuations associated with that filing. Spring 2015 2 UChicago Undergraduate Business Journal The Data Set To determine the impact of publicizing SEC filings on the real-time state of the financial markets, I needed data from the SEC and historical stock market data to compare it against. The SEC Data: The SEC publishes filings through their Edgar search engine and their RSS Feed. I wrote code which copied and parsed the contents of the SEC list of the latest 2,000 filings every second. For each filing this information contained the name of the company, the type of the filing and the time the filing was submitted to the SEC. By comparing the differences between the contents of different scans which were a second or fraction of a second apart, I got a second-resolution timestamp for when the filing was actually posted on the SEC website; if a certain filing is present in the list of most recent 2,000 filings at time T but wasn’t at time T-1 where T and T-1 are one second apart, then there is a one-second window during which the filing might have been uploaded. I collected data on filings from February 2, 2014 to May 8, 2014. I ended up with data on a total of 285,713 filings. The Market Data: From the website Kibot.com, I got ticker data aggregated to one-second intervals for each of the S&P 500 companies. For each interval the data consisted of open, close, high and low values for price and volume. Throughout the rest of this study, I use the open price and the volume quantity to conduct my analysis. Cross-referencing the data Because the SEC filings only had the name of the filer, not a unique identifier of the company to which the filing pertained, I had to make the determination of which company’s stock price and volume movements to analyze around the time of the filing’s posting. Using a conservative word-matching algorithm, I isolated filings whose names contained the name of an S&P 500 company. For example, a Blackrock Inc. filing would be matched to the company Blackrock as would a filing by BlackRock Long-Term Municipal Advantage Trust. Since information about a company’s subsidiaries and branches is directly relevant to the financial value of the parent company’s stock value, I feel that such a matching was reasonable. Spring 2015 3 UChicago Undergraduate Business Journal Types of Filings Of the 285,713 filings analyzed, 2,364 were filed by entities matched to S&P 500 companies. Those 2,364 filings contained a total variety of 77 different types with varying frequencies. The full distribution can be seen in Table 1. Due to time constraints, and out of concern for a sufficiently large sample size, the only types of filings which were analyzed were those whose frequency was greater than 50. Briefly described, they were: 1. 10-K : Comprehensive Annual Report of Company performance 2. 4, 4A : Statement of Changes in Beneficial Ownership of Securities 3. SC 13G/A, SC 13G : Schedule filed to report acquisition of beneficial ownership of 5% or more of a class of equity securities by passive investors and certain institutions 4. 8-K: ‘current report’ of any material current event that happened used to notify investors of any material event that is important to shareholders The 10-K, or Annual Report, contains the most information of any filing. Often spanning hundreds of pages, it contains everything from a letter to shareholders from the board detailing company progress to detailed accounts of the company’s accounting records, including distribution of assets and liabilities, revenues and profits. While it is reasonable to expect that a 10-K would have, by far, the largest impact on prices and volume, this expectation should be tempered by the idea that only new information impacts markets and much of the information in the actual 10-K is made public long before the full 10-K is. In fact, since the 10-K is so long, it is reasonable to expect that it would take more than 5 minutes to analyze it, so there is also reason to expect that there would not be any observable market response to the 10-K within the observed time window, not because 10-Ks are unimportant but because the time window is too short. The 4 or 4A is a notification of legal insider trading. Legally, they are only allowed to trade on public information and so this type of filing should, in theory, contain no relevant information. However, statistically speaking, some of these people might in fact have insider information relevant to the company they work for, and the contents of such a filing might contain important signaling information. An example of a 4A filing can be found in Figure 1. The SC 13G or SC 13G/A is a notification that someone with a large amount of money is interested in purchasing a large percentage of the shares. These filings have a significant implication for market momentum, not to mention the signaling aspect: presumably, if an entity is pouring millions or billions into a financial asset, it has done its research on that asset and has a well-informed understanding of that asset’s true underlying value is relative to the market’s. An example of a SC 13G filing can be found in Figure 2. An 8-K is a catch-all filing for information that the company feels shareholders should know about. An example of an 8-K can be found in Figure 3. However, this is just one example and is not representative of the full range of information that could be contained in an 8-K. Spring 2015 4 UChicago Undergraduate Business Journal While the four types of filings which occurred with a high frequency were highly likely to impact the financial market, the one type of filing which was unfortunately infrequent was the 10-Q: quarterly report of a similar nature to the 10-K but covering a shorter time span and doing so in less detail. I only observed a very small number of 10-Qs (only four, see Table 1) because I collected data over the span of two months, and very few S&P 500 companies defined their company calendar year in a way which resulted in a quarterly report release date during those two months. However, this characteristic of S&P 500 companies is not a major issue because the content of 10-Qs is explicitly a part of the content of 10-Ks. Restricting our data to filings of these types for S&P 500 companies, I was left with a data pool of 1179 4(/A)’s, 446 SC13G(/A)’s, 133 8-K’s, and 51 10-K’s, for a total of 1,809 filings. Mechanism and Time Frame Since equity prices are tied to the real value of the underlying market asset, there are four conditions that must be met for the publishing of SEC filings to have a significant impact on market prices: 1. SEC filings must contain relevant information. I expect this to be the case because the whole point of publishing filings is to make shareholders aware of information they need to know to make informed decisions about the stock. For example, a release of an earnings report which differs from analyst expectations directly affects the implied value of the company; it has less money and will likely make less money in the future. 2. SEC filings must contain new information. Much of the information disclosed in SEC filings is not really new information, at least not to the entire market. For example, the outcome of an election of a new board of directors would be filed with the SEC, but many of the traders who are interested in the outcome of the election would not wait for the outcome to be published on the SEC website and would find out some other way. However, a significant amount of the information in SEC filings is new information to most of the market, like the full contents of an annual report, and could potentially have a significant impact on market prices and volume traded. Additionally, even for information that was likely known, SEC filings provide official confirmation. 3. SEC filings must actually be made public when we think they are. There is the logical possibility that there is insider trading happening on the basis of SEC filings. Potentially, someone in the SEC might be releasing the filing’s contents to certain traders early, before it gets posted on the website, so that they have additional time to react and be the first to trade on that information. If this is really happening, then the observed market reaction to SEC filings would be understated because some or all of the new information would be already priced into the stock’s price when the filing becomes officially posted. Spring 2015 5 UChicago Undergraduate Business Journal 4. Someone must actually be paying attention to the contents of SEC filings and have the capability to process them within the five-minute observation time frame. To isolate the impact of SEC filings from any other new market information, I had to look at an arbitrarily short time window after the filing’s posting to increase the chances that the non-random market changes were related only to SEC filings. I settled on a time window of five minutes. By decreasing our time window to such a small amount, I increased internal validity with respect to which effect I was measuring but greatly compromised our ability to measure its magnitude. Presumably, the entire market doesn’t have an algorithm which monitors the SEC website, quickly scans the content of filings, and makes trades based on the processed information, so the period over which the entire market receives and prices in the contents of a SEC filing could be hours or even days. I was more interested in determining whether SEC filings have an impact, not just in isolating the magnitude of that impact, so I picked five minutes as the length of our post-filing posting time as our observation window because I felt it was a reasonable balance between being long enough for it to be possible for someone to scan through a new SEC filing and make a trade on it and making it short enough that there would likely be no new (or at the very least no non-random) market data relevant to the company associated with any given SEC filing, even if the SEC filing times themselves are non-random. I therefore expect that our findings are a significant understatement of the full impact of the information contained in the SEC filing. Calculating the Variables Having isolated the 1809 S&P 500-affiliated filings of type 4(/A), SC13G(/A), 8-K or 10-K to use in our research, I was ready to match them against historical data. However, the full data set was not usable because the stock exchanges and the SEC’s electronic filing room operate during different hours. The markets are open 9:30 AM to 4:00 PM EST, and the SEC posts filings either at around midnight or from 9:30 AM to 10:00 PM EST. Since it is not possible to observe the effect of a filing’s publishing if the markets are not open when it is published, I was unable to use the majority of our SEC-side data set. I also could not use filings published in the first or last five minutes of the trading day, because it would be impossible to record the market state either before or after the filing. In the end, I was left with 261 filings uploaded within the overlapping 6+2/6th hour time window from 9:35 AM to 3:55 PM EST. This set was further limited by our inability to collect market data on several companies that had only been recently added to the S&P 500. Namely: Facebook, Allegion, Transocean, Grahm Holding Company, Zoetis and Michael Kors (the effect of which was the difference between the frequency values before and after the !). Their breakdown was as follows: Spring 2015 6 UChicago Undergraduate Business Journal • (All): 261 ! 255 • 4: 140 ! 135 • SC 13G(/A): 99 ! 98 • 8-K: 11 ! 11 • 10-K: 11 ! 11 For the filings that were published inside market hours, I used second-resolution aggregated market data to, for each filing, calculate the difference of differences in average stock price and volume change in the five minutes before and after the posting of every filing. For every filing, I then calculated volatility of prices [P] and volume – changes in the quantities traded [V]. To calculate the [P] of any given filing, I took the differences between the stock price when the filing was first posted and the prices five minutes before and after the filing was posted. I then took the absolute value of the difference of the percent change relative to the stock price at posting time. Consider the hypothetical that stocks normally go up +/- 2 percent at equal frequency and after filings they go up +/- 10 percent. In both cases it would average out to a zero percent change, even though filings had a strong impact on volatility. Since I was interested in determining whether filings have an impact on stock prices, not in whether filings generally contain ‘good news’ or ‘bad news’ for the stock, I worked with absolute values. It is important to confirm that I only took the absolute value of the final value. If the stock price went down 5 percent before the filing and went up 5 percent after the filing, that registered as volatility staying flat. To calculate the [V] of any given filing, I calculated the ratio of the total volume in the five minutes after the posting to the total volume in the five minutes prior to the posting, and converted to the relative percent change by subtracting one and then multiplying by 100. Having calculated the [P] and [V] for every filing, I then did the same for semi-random 10-minute snapshots in time. These times were randomly selected within the time constraints of the filings I was using in our data, so they were random 10-minute periods between when I started collecting data in February to when I stopped in May, and were between 9:25 and 3:55. I collected data on 746 such random samples, bringing our total data set up to 1001. I also calculated the time difference between filing time and posting time to determine if this difference has any impact on prices or volume. As discussed, if there is no insider trading and the sample size is sufficiently large, then there should be no observable effect. The distribution of filing-posting time gaps can be found in Figure 6. It was a coincidence that the largest filing-posting gap was just under five minutes, the time interval within which I arbitrarily decided to limit our observations. The Data Spring 2015 7 UChicago Undergraduate Business Journal As a reminder, the breakdown of the filing types can be found in Table 3: Type Frequency Percent Cum No Filing 746 74.53 74.53 4(/A) 135 13.49 88.01 SC 13G(/A) 98 9.79 97.80 8-K 11 1,1 98.90 10-K 11 1.1 100.00 A general variable summary can be found in Table 4: Var Obs Mean Std. Dev. Min Max type 1001 0.408 0.797 0 4 pricediff 1001 0.123 0.122 0 0.946 volumediff 1001 31.39 116.5 -93.5 2404.7 postinggap 1001 10.66034 23.21165 0 294 posted 1001 0.255 0.436 0 1 The distribution of price changes was, as might be expected, the right half of a normal distribution of double height, since I was dealing with absolute values. Histograms of price changes can be found in Figure 4 and Figure 5 on the following page: Spring 2015 8 UChicago Undergraduate Business Journal 0 2 Density 4 6 Figure 4: Price Changes for time intervals including filings 0 .2 .4 .6 .8 Difference in price changes in the five minutes after and before filing 4 0 2 Density 6 8 Figure 5: Price Changes for time intervals not including filings 0 .2 .4 .6 .8 Differences in price changes between the last and first halves of random 10-minute snapshots. 1 The distributions of volume were right-skewed, as expected, since volume can only go down by 100 percent but can go up a potentially infinite amount. They can be found in Figure 6 and Figure 7, also shown on the following page. If you note the different x-axis for Figure 6 and Figure 7, you notice the massive increase in volume change around the time a filing gets posted. Spring 2015 9 UChicago Undergraduate Business Journal 0 .002 Density .004 .006 .008 Figure 6: Percent changes for intervals including filings 0 500 1000 1500 2000 2500 Percent change in volume five minutes after and before posting 0 .002 Density .004 .006 .008 Figure 7: Percent changes in volume for intervals not including filings -200 0 200 400 600 Differences in volume changes between the last and first halves of random 10-minte snapshots Spring 2015 10 UChicago Undergraduate Business Journal .04 0 .02 Density .06 .08 Figure 8: Distribution of filing-posting gaps, in seconds. 0 100 postingGap 200 300 I then regressed both the [P] and [V] on the 0 or 1 posting dummy variable representing whether or not the data came from a 10-minute time window whose five-minute mark was associated with the posting of a filing. Stratified by filing type and then consolidated, our output can be found in Table 5. Table 5: Single Variable Regression, Stratified by Filing Type and Consolidated. Significance level of [P] and [V] labeled [P]*,**, [V]**,***,****. Filing Type [P] [P]*,** [V] [V]**,***,**** Consolidated (All) 0.0091 0.302 36.70 0.000**** 4(/A) (0.0094) 0.405 39.01 0.000**** SC 13G(/A) 0.0289 0.030** 37.01 0.003*** 8-K 0.0074 0.841 57.22 0.028** 10-K 0.0610 0.099* (14.89) 0.558 To clarify, the values in the table are the coefficients and significance of the coefficients of the 0/1 posted dummy variable, where P denotes that volatility of prices [P] was the dependent variable and V denotes that volatility of volume [V] was the dependent variable, with a confidence interval of 95 percent. 0.01:*, 0.05:**, 0.01:***, 0.001**** Lastly, I ran the same regressions as in Table 5, but included the number of seconds between the filing time and posting time as a control for the time associated with a filing. For the random time snapshots, I treated Spring 2015 11 UChicago Undergraduate Business Journal it as missing. Again, a statistically significant non-zero coefficient value for the control would be evidence of insider trading. The results can be found in Table 6. Table 6: Regression using time to post filing as explanatory variable, consolidated. Filing Type [P] [P]*,** [V] [V]**,***,**** Consolidated 0.0018 0.502 (0.3863) 0.326 The significance level of [P] and [V] are labeled [P]*,** and [V]**,***,****. To clarify, the values in the table are the regression coefficients and significance levels of two of the independent variables: the posting gap (number of seconds between posting and filing) P and the volume (V). The regressions in this table do not include the randomly sampled 10-minute segments which were not associated with a timestamp. Discussion and Interpretation The objectives of this study were threefold: 1. Determine if SEC filings are noticed in the market and, if so, what types? 2. Measure the extent to which they have a real-time impact. 3. Determine whether the early dissemination of SEC filing data is a mechanism for insider trading. Are SEC filings noticed by the market? Yes. The key underlying assumption that we use to eliminate the need for controls is that, given a small enough time frame and a potentially market-impacting event in the middle of that time frame, any non-random fluctuations in price or volume in the period after that event relative to the period before can be causally attributed to that event. Since I limited my observation window to five minutes before and after the event, a filing’s posting, I am reasonably confident that there is sufficient internal validity to conclude that the variations we observed are directly attributable to the SEC filing. Therefore, since there is an observed increase in volume, significant to p<0.001, I conclude that SEC filings are monitored by trading entities in the market who respond to their contents within the first five minutes of their publication. How impactful are SEC filings, really? Unknown. While the five-minute post-filing posting observation window increases internal validity, it greatly limits the potential to observe the full effect of the SEC filings because not everyone in the market is responding to SEC filings within five minutes. One reason that this is the case is that some types of filings take much longer than five minutes to analyze. For example, 10-Ks are hundreds of pages long, and can take days or even weeks to interpret. This is likely why, when stratified, 10-Ks were the only filing type which did not have a statistically significant impact on volume. Spring 2015 12 UChicago Undergraduate Business Journal For this reason, the price data is unable to accurately represent the financial implication of a SEC filing because the full effect of the filing’s posting takes place over a long period of time, and initially there are always actors willing to provide liquidity in the market, essentially creating a buffer zone. Thus the initial trades based on SEC filings are unlikely to move the price, and only after many people make trades based on the data will the market actually meet the demand by increasing price. Of course, active market makers significantly increase their bid-ask spread when they notice a change in demand. The actual buffer zone is created by inactive customers through standing limit orders, a type of order which is placed and waits until prices sufficiently move (Sarkar, Locke 1999). If there are enough standing limit orders to purchase a penny above the market price, the price can’t move more than a penny until all those orders are satisfied. Finally, price is an insufficient indicator of market response to filings because the contents of SEC filings are not unambiguously good or bad. They may contain a mixed bag of information that results in lots of trades one way or the other which average out to zero and are not reflected in a price change. On the other hand, because volume has no such drawbacks and is generally low absent new market information, it is much easier to observe the impact of SEC filings through fluctuations in volume. The increase in volume of 36.70 percent in the first five minutes after a filing might intuitively seem large, but there’s no way to know whether that is because there’s little activity absent information or because the new information contained in SEC filings is particularly interesting to the market. The best way to determine the impact of SEC filings would be to quantify it in dollars by multiplying the change in market price by the number of shares outstanding. Again though, that would need to be calculated over the full period during which people make trades based on the SEC filings. Past the time frame of fiveminutes, it becomes harder to attribute market price fluctuations solely to filing-related new information, so without actually controlling for other types of new information, a data set in general is difficult to obtain and certainly inaccessible to me. Thus it is not possible to quantify the financial impact of an SEC filing at this time. However, based on the relative volume fluctuations observed through stratifying by filing type, we can make a relative observation about which filings tend to be more important. The 4(/A) and SC 13G(/A) type filings resulted in a volume increase of 35-40 percent, but the 8-Ks resulted in a relatively massive increase of almost 60 percent, suggesting that the information contained in the 8-Ks is likely to have a significantly greater impact on the market. This relationship provides interesting insight into what type of information the market considers important. As a reminder, 4(/A) type filings publicize the fact that someone inside the company purchased or sold some of the company’s stock, and SC 13G(/A) type filings publicize that someone outside the company made a move to acquire five percent or more of the entire stock. So these filings provide a strong signal about the changing opinions of individual actors in the market, but no new information. However, 8-Ks are material notices to shareholders about recent developments and do contain new information. The relatively higher increase in volume associated with 8-Ks is an indication that entities in the markets are more interested in Spring 2015 13 UChicago Undergraduate Business Journal tangible new information than in signals provided by other entities making their decision based on already existing information. This interpretation is also consistent with the fact that 4(/A) filings have a slightly higher impact on volume than SC 13G(/A)’s. Large institutions purchasing more than five percent of the stock do a lot of research into the equity, so a SC 13G(/A) represents a lot of research based on current market knowledge. On the other hand, a 4(/A) represents potentially no research but may represent a small amount of insider knowledge since it is a trade made by someone inside the company. It is quite interesting that the markets consider a potential unknowable quantity of insider knowledge more important than the fruits of significant due diligence. Is early dissemination of SEC filings resulting in insider trading? Maybe. Insider trading is likely being done by a relatively small portion of the market, so the effect of insider trading, or early market reaction to an SEC filing, would have a much smaller magnitude than the rest of the market responding to the SEC filing. Also, the direction of the impact of insider trading would be opposite the direction of the response of the rest of the market, because the market response after the filing’s posting was measured relative to that of the market state before the posting, and if the real market response actually began earlier then it would partially cancel itself out and the observed effect would be weakened. Both of these conditions were observed when using volume as the dependent variable. It was of small magnitude and negative, both of which would be predicted if insider trading was occurring. Using the dependent variable, the price change was almost zero but actually positive. This cannot be clearly interpreted as evidence for or against insider trading because there are two effects going on that determine the sign of the relative absolute volatility. First, insider trading would work to cancel out the later market response, increasing the chance of a decrease in observed price fluctuations after posting. This would result in a negative sign. Second, insider trading would break down the liquidity buffer in advance of the posting, increasing volatility and making prices more sensitive to the rest of the market response to follow. This would result in a positive sign. The observed price magnitude fluctuation was 0.0018 percent which is so close to zero that, even if we assume insider trading, we cannot determine which of these two effects are stronger. So the volume-regression is the only information we can use to determine whether there is insider trading. While both the sign and magnitude of the observed fluctuations are indicative of insider trading, these were found to be statistically significant only to the p<35 percent level. Absent the sign and magnitude coincidence, this would be no evidence at all, but in conjunction I would argue that there is weak evidence that insider trading is happening. Spring 2015 14 UChicago Undergraduate Business Journal Discussion of Limitations and Potential Sources of Error There are several potential sources of error that may have compromised either internal or external validity: 1. Accuracy of mapping filing to companies 2. Representativeness of the counterfactual sample 3. Insufficiently small time window of observation around filing time 4. Insufficiently large time window of filing data set collection time 5. Lack of extrapolation across filing types and companies Accuracy of mapping filing to companies As explained earlier, there was no direct one-to-one mapping from filing name to company stock symbol, so I used a word matching algorithm which produced the aggregate contents of Table 2. The actual implementation of the algorithm can be found in the code in the appendix, but the basic procedure was to strip the filing name of punctuation and find a company whose name was contained in the filing name and then to match the name to the stock symbol of that company. Subjectively, both the algorithm and the output seem reasonable, but it is possible that this procedure could result in an underestimate of the magnitude of the filing’s impact. If the algorithm were less than perfect, then some filings would be matched with companies they were unrelated to and would barely, if at all, affect the real-time price or volume traded of the company’s stock. Representativeness of the counterfactual sample To determine the ‘normal’ price and volume fluctuation of a typical stock in our sample, we used random 10-minute snapshots which were unrelated to a filing. I could not rely entirely on the data obtained in the five minutes before the filing as the counterfactual because, while it would provide a great deal of useful information about the time and date of the observed filings the data could be affected by insider trading in a way that would be difficult to later decompose. Still, to be a counterfactual, the random sample needed to be constrained in such a way that the sample of random times would be representative of the times in which we observed filings and the companies against which we matched them. To provide a counterfactual with respect to the time window, we randomly sampled 10-minute time segments within the set of dates during which we were collecting data and within market hours, in a manner similar to how we obtained our primary data set. There are several potential issues with this method. First, the code occasionally threw an error and stopped the data collection (in a way that did not impact the accuracy of already-collected data), causing there to be intermittent gaps of hours or in some cases even several days during which data collection stopped. Second, SEC filing posting times are non-random; they are posted soon after they are filed, and filing times are non-random. However, this is not a large issue of concern because market trends Spring 2015 15 UChicago Undergraduate Business Journal that might be true during some parts of the day but not others are tiny compared to the patterns we observed in other data. Still, this was a variable that was not fully controlled for. For a counterfactual with respect to the companies against whose stock we matched the random time sampled (which was random, and not associated with a filing, and so had no obvious company to match it to), we picked a randomly representative sample of the set of companies whose filings constituted the original data set. That is, we only matched the random timestamp against a company to which we matched at least one filing, and the likelihood of picking that company was directly proportional to the fraction of filings in the total data set matched to that company; if company A was matched to by eight out of the 255 filings, it had an eight in 255 chance of having its stock’s historical data be examined around the time of any random counterfactual timestamp. Insufficiently small time window of observation around filing time. As already mentioned, the length of the time-window of historical market data around a filing’s posting time had to be kept small to avoid the need for controlling for the introduction of new market information. Given a small enough time window, the non-random difference in the likelihood of there being new market data unrelated to the SEC filing immediately before and after its posting would be 0, even if the times when SEC filings are filed are non-random. I believe that our five-minute window before and after, or 10-minutes total, is sufficiently small to ensure that this is the case. This is an assumption that if false would be the source of error. Still, it is not a major concern for the same reason that the random time sample not being a perfect counterfactual with respect to time isn’t a major concern: market fluctuations correlated with the date are much smaller than those observed in our data. Insufficiently large time window of filing data set collection time. Since I only collected data on filings for several months, there is insufficient data to observe seasonal patterns or relationships between many types of filings. For example, if 4(/A)s have more of an impact during the first and third fiscal quarters, but 8-Ks are more meaningful in the second and fourth (for whatever reason), that and other potentially interesting patterns have gone unobserved. More importantly, a larger data set would result in a more statistically significant, and thus more definitive, answer to the question of whether or not there are bad actors within the SEC who facilitate insider trading. Lack of generalizability across filing types and companies Given the limited scope and size of the data set, I am unable to apply my claims about the market reaction to certain types of filings by S&P 500 companies to other filing types, time periods, or companies. The S&P 500 index was chosen because it was an arbitrary set of large, recognizable companies and was thus easier to work with. However, since it is a non-random set, it is quite plausible that the markets largely ignore SEC filings for smaller companies, or even that markets pay even more attention to filings for smaller companies because that’s the main source of information about the company. I can read a press release about Microsoft’s earnings, Spring 2015 16 UChicago Undergraduate Business Journal but if I want any important information about the other, smaller tech company Miosoft’s earnings, I would have to actually read their 10-K. Similar concerns extend to filing types, which again were selectively picked by frequency, familiarity and import, all factors which affect the extent to which the market pays attention to them. Conclusion The primary objective of this paper was to establish that SEC filings impact stock prices or volume immediately after publication, a claim which is well supported by our evidence. With a confidence level of p < 0.001, SEC filings increase volume within five minutes of publication. For two types of filings, SC 13G(/A) and 10-K, I also found statistically significant movements in price within the same time frame. These findings show both that SEC filings contain new, market-relevant information and that there are actors in the market actively monitoring the SEC and making trades based on that information within a short time frame. The secondary objectives were to quantify the magnitude of the financial impact of SEC filings and to discover which types of filings have the greatest impact. However, to isolate the impact of SEC filings from other potential sources of market information, I had to restrict my observations to a time window of five minutes, much shorter than the period of time over which the markets respond to the contents of SEC filings. As a result, I was unable to quantify to cumulative impact of SEC filings. By stratifying my regressions by filing type, I did find an interesting pattern: the filing types which resulted in the greatest volume increases were those based on new market information, rather than those based on signals from large players in the market acting on previously current market information. A tertiary but important objective was to determine if there was any evidence that insider trading might be facilitated through the early release of SEC filings. I predicted that if insider trading was occurring, I would observe a small negative coefficient when regressing the time it takes the SEC filings on the observed change in volume. This follows from the theory that insider trading would shift the market response to the filing backwards in time before the official posting of the filings, reducing the observed magnitude of the market’s response. While this is exactly what was found in the data, the predicted coefficient was not statistically significant, so at best I found weak evidence of insider trading. Given the relative dearth of academic research and data collection on SEC filings in comparison to other areas of trading and finance, there is still substantial work to be done in this area. Specifically, a larger data set should be collected and analyzed so that patterns related to more filing types, or to alternative types of stratifications such as by industry sector, could be observed. Spring 2015 17 UChicago Undergraduate Business Journal Appendix 1. Tables Table 1: Filing Types, sorted by frequency Filing Type 4 SC 13G/A 8-K SC 13G 497K 10-K 497 10-D 3 13F-NT 485BPOS FWP DEFA14A DEF 14A 424B5 N-CSR S-8 POS N-Q 5 ARS 4/A NSAR-B Freq. (Max-11) 1167 346 133 100 53 51 43 34 33 31 28 22 19 18 16 15 15 14 13 13 12 11 Filing Type 8-K/A 13F-HR (Filer) 13F-NT NT 10-K D/A EFFECT PRE 14A 424B3 ABS-15G SC 13D/A 144 NO ACT 424B2 SC TO-I 40-6B/A 485APOS N-1A/A 10-K/A (Filer) D/A POS 8C SC TO-I/A 10-Q D S-3ASR N-CSRS 8-A12B 497J SC 13D NSAR-A Freq. (10-3) 10 10 9 9 7 7 6 6 5 5 5 5 5 4 4 4 4 4 4 4 4 4 3 3 3 3 3 3 3 Filing Type POSASR CERTNYS TA-2 24F-2NT 40-APP 10-Q/A ABS-15G/A (Filer) 13F-HR 13F-HR/A PREM14A 15-15D 15-12G CERTNAS DEFR14A IRANNOTICE PRER14A 3/A NCSR/A (Filer) 40-6B/A 25-NSE 40-17G S-8 CT ORDER POS AM N-8F 10-D Freq. (2-1) 2 2 2 2 2 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 Spring 2015 18 UChicago Undergraduate Business Journal Table 2: Mappings from Filing Company Name to Stock Symbol Filing Name Stock Mapping Frequency Allegion plc ALLE 1 Avery Dennison Corp AVY 3 BlackRock Alternatives Allocation Master Portfolio LLC BLK 1 BlackRock EcoSolutions Investment Trust BLK 1 BlackRock Energy & Resources Trust BLK 3 BlackRock Enhanced Equity Dividend Trust BLK 2 BlackRock Enhanced Government Fund, Inc. BLK 1 BlackRock Fund Advisors BLK 1 BlackRock Health Sciences Trust BLK 3 BlackRock Inc. BLK 25 BlackRock International Growth & Income Trust BLK 2 BlackRock Long-Term Municipal Advantage Trust BLK 1 BlackRock Preferred Partners LLC BLK 1 BlackRock Real Asset Equity Trust BLK 1 Delphi Automotive PLC DLPH Discover Financial Services DFS 1 Discovery Communications, Inc. DISCA 1 Dr Pepper Snapple Group, Inc. DPS 1 Duke Energy CORP DUK 1 Express Scripts Holding Co. ESRX 1 Fidelity National Information Services, Inc. FIS 3 Fixed Income Trust for Prudential Financial, Inc. Notes, Series 2012-1 PRU 1 GameStop Corp. GME 1 General Mills Group Trust GIS 1 Google Inc. GOOG 8 Graham Holdings Co GHC 1 Invesco Ltd. IVZ 38 Legg Mason Investment Counsel, LLC LM 1 LyondellBasell Industries N.V. LYB 1 Marathon Petroleum Corp MPC 13 Mead Johnson Nutrition Co MJN 11 Merck & Co. Inc. MRK 13 Michael Kors Holdings Ltd KORS 1 Monster Beverage Corp MNST 2 20 Spring 2015 19 UChicago Undergraduate Business Journal Filing Name (cont’d) Stock Mapping Frequency Morgan Stanley Bank of America Merrill Lynch Trust 2012-C6 MS 1 Morgan Stanley Bank of America Merrill Lynch Trust 2013-C10 MS 1 Morgan Stanley Bank of America Merrill Lynch Trust 2013-C11 MS 1 Morgan Stanley Bank of America Merrill Lynch Trust 2013-C12 MS 1 Morgan Stanley Bank of America Merrill Lynch Trust 2013-C13 MS 1 Morgan Stanley Bank of America Merrill Lynch Trust 2013-C9 MS 1 Morgan Stanley Capital I Trust 2012-C4 MS 1 Noble Corp plc NE 5 Nordstrom James F JR JWN 1 Philip Morris International Inc. PM 11 Public Storage PSA 9 Sabra Health Care REIT, Inc. HCN 5 Spectra Energy Corp. SE 11 TE Connectivity Ltd. TEL 23 Transocean Ltd. RIG 4 TripAdvisor, Inc. TRIP 1 Unum Group UNM 1 Viacom Inc. VIAB 2 Vulcan Materials CO VMC 3 Wells Fargo Commercial Mortgage Trust 2013-LC12 WFC 1 Xylem Inc. XYL 9 Zoetis Inc. ZTS 2 Table 3: Breakdown of the filing types Type No Filing 4(/A) SC 13G(/A) 8-K 10-K Frequency 746 135 98 11 11 Percent 74.53 13.49 9.79 1,1 1.1 Cum 74.53 88.01 97.80 98.90 100.00 Spring 2015 20 UChicago Undergraduate Business Journal Table 4: Variable Summary Var type pricediff volumediff postinggap posted Obs 1001 1001 1001 1001 1001 Mean 0.408 0.123 31.39 10.66034 0.255 Std. Dev. 0.797 0.122 116.5 23.21165 0.436 Min 0 0 -93.5 0 0 Max 4 0.946 2404.7 294 1 Table 5: Single Variable Regression, Stratified by Filing Type and Consolidated. Filing Type Consolidated (All) 4(/A) SC 13G(/A) 8-K 10-K [P] 0.0091 (0.0094) 0.0289 0.0074 0.0610 [P]*,** 0.302 0.405 0.030** 0.841 0.099* [V] 36.70 39.01 37.01 57.22 (14.89) [V]**,***,**** 0.000**** 0.000**** 0.003*** 0.028** 0.558 Significance level of price [P] and volume [V] labeled [P]*,**, [V]**,***,****. All regressions have a confidence interval of 95%. 0.01:*, 0.05:**, 0.01:***, 0.001**** Table 6: Regression using time to post filing as explanatory variable, consolidated. Filing Type Consolidated [P] 0.0018 [P]*,** 0.502 [V] (0.3863) [V]**,***,**** 0.326 The significance level of [P] and [V] are labeled [P]*,** and [V]**,***,****. To clarify, the values in the table are the regression coefficients and significance levels of two of the independent variables: the posting gap (number of seconds between posting and filing) P and the volume (V). The regressions in this table do not include the randomly sampled 10-minute segments which were not associated with a timestamp. Spring 2015 21 UChicago Undergraduate Business Journal 2. Figures Figure 1. Example of SEC Form 4 SEC Form 4 FORM 4 OMB APPROVAL UNITED STATES SECURITIES AND EXCHANGE COMMISSION OMB Washington, D.C. 20549 Number: 3235-0287 Check this box if no December longer subject to Section 16. Form 4 or Form 5 Expires: STATEMENT OF CHANGES IN BENEFICIAL OWNERSHIP 31, 2014 obligations may Estimated average continue. See Instruction Filed pursuant to Section 16(a) of the Securities Exchange Act of 1934 1(b). or Section 30(h) of the Investment Company Act of 1940 burden hours per 0.5 response: 1. Name and Address of Reporting Person* 2. Issuer Name and Ticker or Trading Symbol 5. Relationship of Reporting Person(s) to Lawson Douglas A. AXCELIS TECHNOLOGIES INC [ ACLS ] Issuer (Check all applicable) 3. Date of Earliest Transaction (Last) (First) (Middle) C/O AXCELIS TECHNOLOGIES, INC. Director (Month/Day/Year) Other 05/09/2014 Officer (give X 108 CHERRY HILL DRIVE 10% Owner (specify title below) below) EVP, Corporate Mktg & Strategy (Street) BEVERLY (City) MA (State) 01915 (Zip) 4. If Amendment, Date of Original Filed 6. Individual or Joint/Group Filing (Check (Month/Day/Year) Applicable Line) 05/12/2014 Form filed by One Reporting X Person Form filed by More than One Reporting Person Table I - Non-Derivative Securities Acquired, Disposed of, or Beneficially Owned Spring 2015 22 UChicago Undergraduate Business Journal 1. Title of 2. Transaction 2A. Deemed 3. Security Execution Date, if Transaction (A) or Disposed Of Date (Instr. 3) (Month/Day/Year) any (Month/Day/Year) 4. Securities Acquired 5. Amount of 6. 7. Nature of Securities Ownership Indirect Code (Instr. (D) (Instr. 3, 4 and 5) Beneficially Form: Beneficial 8) Owned Direct (D) Ownership Following or Indirect (Instr. 4) Reported (I) (Instr. 4) Transaction(s) (Instr. 3 and 4) (A) Code V Amount or Price (D) Common 05/09/2014 P 528 A $1.58 10,626 D 05/09/2014 P 4,472 A $1.59 15,098 D 05/09/2014 P 5,000 A $1.59 12,500 I Stock Common Stock Common By spouse Stock Spring 2015 23 UChicago Undergraduate Business Journal Figure 2. Example of a SC 13G filing (Parts without relevant information have been omitted). SC 13G/A 1 americanpubliceduinc04302014.htm AMERICAN PUBLIC EDUCATION INC. SECURITIES AND EXCHANGE COMMISSION Washington, D.C. 20549 SCHEDULE 13G/A (Rule 13d-102) INFORMATION TO BE INCLUDED IN STATEMENTS FILED PURSUANT TO RULES 13D-1(b) AND AMENDMENTS FILED THERETO FILED PURSUANT TO RULE 13D-2(b) Under the Securities Exchange Act of 1934 (Amendment No. 2)* American Public Education Inc. (Name of Issuer) Common Stock (Title of Classes of Securities) 02913V103 (CUSIP Number) April 30, 2014 (Date of Event Which Requires Filing of this Statement) CUSIP No.:02913V103 1 NAME OF REPORTING PERSON I.R.S. IDENTIFICATION NO. OF ABOVE PERSON (ENTITIES ONLY) Invesco Ltd. IRS # 980557567 4 CITIZENSHIP OR PLACE OF ORGANIZATION Invesco Ltd. – Bermuda NUMBER OF SHARES BENEFICIALLY OWNED BY EACH REPORTING PERSON WITH 9 5 SOLE VOTING POWER – 1,771,324 6 SHARED VOTING POWER – 0 7 SOLE DISPOSITIVE POWER – 1,771,324 8 SHARED DISPOSITIVE POWER – 0 AGGREGATE AMOUNT BENEFICIALLY OWNED BY EACH REPORTING PERSON 1,771,324 Item 1(a). Name of Issuer: American Public Education Inc. Spring 2015 24 UChicago Undergraduate Business Journal Figure 2, cont’d (b). Address of Issuer’s Principal Executive Offices: 111 West Congress Street; Charles Town, WV 25414;United States Item 2(a). Name of Person Filing: Invesco Ltd. (b). Address of Principal Business Office or, if none, residence of filing person: 1555 Peachtree Street NE; Atlanta, GA 30309; United States (c). Citizenship of filing person: Bermuda (d). Title of Classes of Securities: Common Stock (e). CUSIP Number: 02913V103 Item 3. If this statement is filed pursuant to ss240.13d-1(b) or 240.13d-2(b) or (c), check whether the person filing is a: (e) [x] An investment adviser in accordance with section 240.13d-1(b)(1)(ii)(E) (g) [x] A parent holding company or control person in accordance with section 240.13d-1(b)(1)(ii)(G) Item 6. Ownership of More than Five Percent on Behalf of Another Person: Invesco Canada Ltd. is subsidiary of Invesco Ltd. and it advises the Invesco Select Companies Fund which owns 8.14% of the security reported here in. However no one individual has greater than 5% economic ownership. The shareholders of the Fund have the right to receive or the power to direct the receipt of dividends and proceeds from the sale of securities listed above. Item 7. Identification and Classification of the Subsidiary which Acquired the Security Being Reported on by the Parent Holding Company: The following subsidiaries of Invesco Ltd. are investment advisers which hold shares of the security being reported: Invesco Canada Ltd. Invesco Advisers, Inc. Invesco PowerShares Capital Management Spring 2015 25 UChicago Undergraduate Business Journal Figure 3. Example of n 8-K filing (Parts without relevant information have been omitted). CURRENT REPORT Pursuant to Section 13 or 15(d) of the Securities Exchange Act of 1934 Date of Report (Date of Earliest Event Reported): May 7, 2014 Aqua America, Inc. (Exact name of registrant as specified in its charter) Pennsylvania 001-06659 23-1702594 (State or other jurisdiction (Commission (I.R.S. Employer of incorporation) File Number) Identification No.) 762 West Lancaster Avenue, Bryn Mawr, Pennsylvania 19010-3489 (Address of principal executive offices) (Zip Code) Item 5.07 Submission of Matters to a Vote of Security Holders The 2014 Annual Meeting of Shareholders of Aqua America, Inc. (the “Company”) was held on May 7, 2014 at the Drexelbrook Banquet Facility & Corporate Events Center, 4700 Drexelbrook Drive, Drexel Hill, Pennsylvania 19026, pursuant to the Notice sent, beginning on March 27, 2014, to all shareholders of record at the close of business on March 10, 2014. At the annual meeting: 1. The following nominees were elected as directors of Aqua America, Inc. to serve for one-year terms and until their successors are elected and qualified. The votes received are set forth adjacent to the names below: Name of Nominee Nicholas DeBenedictis For Withheld 97,519,584 6,675,058 Michael L. Browne 102,251,073 1,943,569 Richard H. Glanton 101,168,228 3,026,414 97,227,613 6,967,029 William P. Hankowsky 102,464,622 1,730,020 Wendell F. Holland 102,799,101 1,395,541 Ellen T. Ruff 102,659,968 1,534,674 Andrew J. Sordoni, III 102,482,214 1,712,428 Lon R. Greenberg There were a total of 41,134,185 broker non-votes for the election of directors. Spring 2015 26 UChicago Undergraduate Business Journal Figure 3, cont’d 2. The appointment of PricewaterhouseCoopers LLP as the independent registered public accounting firm for the Company for the year ending December 31, 2014, was ratified by the following vote of shareholders: For Against Abstain 142,897,940 1,694,427 736,460 3. The advisory vote to approve the Company’s executive compensation as disclosed in the Company’s Proxy Statement for the 2014 Annual Meeting of Shareholders was approved by the following vote of shareholders: For Against Abstain Broker Non-Votes 96,232,744 5,629,928 2,331,970 41,134,185 4. The Amended Aqua America, Inc. 2009 Omnibus Compensation Plan was approved by the following vote of shareholders: For Against Abstain Broker Non-Votes 97,668,672 4,830,738 1,695,232 41,134,185 5. The shareholder proposal requesting that the Board of Directors create a comprehensive policy articulating the Company’s respect for and commitment to the human right to water was not approved by the shareholders and received the following vote: For Against Abstain Broker Non-Votes 11,177,742 88,408,849 4,608,051 41,134,185 6. The shareholder proposal requesting that the Board of Directors create a policy in which the Board’s Chairman is an independent director who has not previously serves as an executive officer of the Company was not approved by the shareholders and received the following vote: For Against Abstain Broker Non-Votes 36,366,308 66,289,197 1,539,137 41,134,185 Spring 2015 27 UChicago Undergraduate Business Journal 0 2 Density 4 6 Figure 4: Price Changes for time intervals including filings 0 .2 .4 .6 .8 Difference in price changes in the five minutes after and before filing Figure 5: Price Changes for time intervals not Figure 6: Percent changes for intervals including Percent change in volume five minutes after and before posting 2000 1500 1000 500 0 Density .004 .006 .008 .002 0 filings 0 .2 .4 .6 .8 1 Differences in price changes between the last and first halves of random 10-minute snapshots. 2500 including filings 0 2 4 Density Spring 2015 28 6 8 UChicago Undergraduate Business Journal seconds. 200 400 200 postingGap 100 0 -200 0 300 not including filings 600 Figure 8: Distribution of filing-posting gaps, in Differences in volume changes between the last and first halves of random 10-minte snapshots Figure 7: Percent changes in volume for intervals 0 .02 .04 Density Spring 2015 29 .06 .08 0 .002 Density .004 .006 .008 UChicago Undergraduate Business Journal Works Cited Feroz, Ehsan H. and Park, Kyung Joo and Pastena, Victor, The Financial and Market Effects of the SEC's Accounting and Auditing Enforcement Releases (July 24, 2008). Journal of Accounting Research, Vol. 29, pp. 107-142, 1991. Available at SSRN: http://ssrn.com/abstract=1175102 Darren T. Roulstone (1999) Effect of SEC Financial Reporting Release No. 48 on Derivative and Market Risk Disclosures. Accounting Horizons: December 1999, Vol. 13, No. 4, pp. 343-363. doi: http://dx.doi.org/10.2308/acch.1999.13.4.343 Thomas J. Linsmeier, Daniel B. Thornton, Mohan Venkatachalam, and Michael Welker (2002) The Effect of Mandated Market Risk Disclosures on Trading Volume Sensitivity to Interest Rate, Exchange Rate, and Commodity Price Movements. The Accounting Review: April 2002, Vol. 77, No. 2, pp. 343-377. doi: http://dx.doi.org/10.2308/accr.2002.77.2.343 Peter R. Locke (1999), Commodity Futures Trading Commission George Washington University; Liquidity Supply and Volatility: Futures Market Evidence http://www.newyorkfed.org/research/economists/sarkar/sarkar_locke.pdf Spring 2015 30