Can Patent Filing Rates Predict Stock Prices?
We hypothesized that the patent application filing rates for typical public technology companies are correlated with their stock prices. This hypothesis was premised on three assumptions:
1. There is a strong correlation between R&D investment and patent filings for a typical technology company;
2. Companies adjust R&D spending quickly in response to economic changes in market conditions; and
3. In an efficient market, changes in economic conditions will directly or indirectly trigger changes in stock prices.
Consequently, we conjectured that the patent filing rate for a typical public company would be strongly correlated with its stock price, and we wanted to test this premise.
To push further this analysis, we wanted to explore the causality between the patent filing rates and stock prices. We added one additional assumption:
4. Adjustments in internal financial forecasts will lead a typical technology company to promptly adjust its R&D spending, and therefore its patent filing rate, before the information about its adjusted economic outlook reaches the external market.
Based on this additional assumption, we hypothesized that the patent filing rate of a typical technology company could be used to predict its stock price in advance. We wondered if we could test this hypothesis, and if so, we wondered what the correlation lead time could be.
Each of these four assumptions is inaccurate at some level, but they should not by themselves fully negate the hypotheses.
Overview of Results
Our results confirmed systematic correlation between the patent filing rates of technology companies active in the Commerce space and their stock prices. Our analysis further indicated that the patent filing rates could be used to predict stock prices with a lead of ~30 days, which should be good news for our friends in the public markets investment community.
We also observed a correlation factor that suggested a parallel reverse causality, with the patent filing rate also appearing to respond to economic conditions from the prior 30 days.
Both of these conclusions are discussed in more detail below.
Summary of Analysis
To test these hypotheses, we initially avoided large conglomerates with diversified operations across multiple industry segments (e.g., GE, HP, Intel) because we thought that their stock prices could be somewhat immune to immediate economic variations and that such companies would not necessarily adjust R&D spending immediately in response to economic changes. We also avoided companies with long-term global patent filing strategies (e.g., IBM, Samsung) because we assumed that their patent filing rates would not be adjusted promptly in response to economic market changes.
We focused on companies active in the Commerce space, including FinTech, Payments, Omnichannel and Data Analytics, and we selected only patents related to Commerce. We retrieved adjusted stock prices automatically from Yahoo Finance, using Yahoo’s API and the pandas_datareader Python library. We limited the timeframe of the analysis to the period between January 2001 and December 2014 for two reasons: (a) we wanted to avoid the stock bubble and subsequent deflation around 2000, which occurred before Commerce patent filings ramped up in the US and globally, and (b) we wanted to ensure that all applications filed recently had sufficient time to become public and to be reported by the various worldwide Patent Offices to maximize the patent application dataset. We considered global patent filings for the companies of interest (as opposed to just US patent applications) with the assumption that foreign patent applications will typically originate from US R&D budgeting for US public companies and also to avoid having to segregate R&D spending across geographies.
Fig. 1 shows the number of patent applications filed by MasterCard on the left axis v. MasterCard’s adjusted stock price on the right axis. The Stock price is shown as 0 until May 2006, when MasterCard filed its IPO and went public. The data had significant variability at a monthly level, so we used a running prior 3-month average for both data series. Please see the analysis towards the end of this article for a more rigorous analysis that computes the Pearson Correlation Coefficient for a dataset at the monthly-level.
After May 2006, Figure 1 shows a clear correlation between the number of patent applications filed by MasterCard and its stock price. We note some volatility in the number of applications filed even with a running 3-month average. This is not unusual for most companies because patents often tend to be filed in families, in response to discrete R&D projects, or are sometimes expedited through the application process in response to upcoming time bars.
Fig. 1: MasterCard Commerce Patent Applications Filed v. Adjusted Stock Price
(3-Months Running Average)
Fig. 2 shows a similar analysis for Amazon, with the number of patent applications filed by Amazon globally plotted on the left axis and Amazon’s adjusted stock price on the right axis (prior 3-months running average for both data series again). As seen in Figure 2, there is again a clear correlation between the number of patent applications filed and the stock price, stronger than in Fig. 1.
Fig. 2: Amazon Commerce Patent Applications Filed v. Adjusted Stock Price
(3-Months Running Average)
Fig. 3 shows the number of patent applications filed by Facebook globally on the left axis v. Facebook’s adjusted stock price on the right axis (again both with a 3-month running average to help smooth out short term variations). Facebook started to trade publicly in May 2012, and subsequently the stock price and the number of patent applications appear to be clearly correlated, despite ongoing volatility in the patent application filings.
Fig. 3: Facebook Commerce Patent Applications Filed v. Adjusted Stock Price
(3-Months Running Average)
Figs. 1-3 show a significant degree of correlation between the number of applications filed on a monthly basis and the stock price for MasterCard, Amazon and respectively Facebook, which is validating our initial hypothesis. We expected to see strong correlation, but did not expect to see a perfect alignment due to a number of factors, including (a) some inherent uncertainty in the underlying set of assumptions, (b) we are considering patent datasets limited to Commerce while comparing to the stock price, which is more indicative of an overall company business performance, and (c) individual companies are likely subject to more exogenous variability. And despite these limiting factors, we indeed observed a remarkable degree of correlation between patent filings and stock prices for these companies.
It is intriguing therefore to test this correlation hypothesis on a special patent-filing company like IBM. We deferred IBM's dataset initially because we assumed that IBM's long term patent filing strategy would be decoupled to a certain extent from short-term economic variability. Fig. 4 shows the patent application and stock price for IBM for the same timeframe:
Fig. 4: IBM Commerce Patent Applications Filed v. Adjusted Stock Price
(3-Months Running Average)
It is interesting to note that even for IBM, the number of applications filed tends to correlate with the stock price at a macro level, despite the patent filings overshooting during 2008 and then lagging starting in 2009. But clearly the correlation is much lower than for the companies with a higher focus on Commerce and less focus on strategic patent portfolio development.
We extended our hypothesis and postulated that at a broader industry level, the correlation between the applications filed and stock prices would hold more accurately even when including companies like IBM. The reason behind this assumption was that averaging over more companies will further attenuate higher frequency noise in individual signals, leading to a statistical averaging over both the stock prices and patenting rates.
Fig. 5 shows a compounded graph comparing the total number of patent applications filed by a set of 26 companies with activities in Commerce v. the compounded stock prices for all of those companies (considering past 3-month running averages again to reduce short-term volatility):
Fig. 5: Select Commerce Companies: Commerce Patent Applications Filed v. Adjusted Stock Price
(3-Months Running Average)
The following set of 26 companies were included in the analysis of Fig. 5: Alibaba, Amazon, American Express, Apple, Bank of America Merrill Lynch (BOAML), Cisco, Citibank, eBay, Facebook, Google, HP, IBM, Intel, JP Morgan Chase, MasterCard, Microsoft, Oracle, PayPal, SAP, Sony, Square, Target, Visa, Walmart, Wells Fargo and Yahoo.
As shown in Fig. 5, the compounded number of patent applications and stock prices of these 26 companies are clearly correlated, despite some volatility in the two curves. Further, as we suspected, the larger dataset for a broader index of companies significantly increases the degree of correlation compared to the analysis for individual companies in Figs. 1-4.
We reproduced the analysis from Fig. 5 again for Fig. 6, but this time comparing the total number of patent applications filed by the companies considered in Fig. 5 against the full S&P 500 Index (using prior 3-month running averages again):
Fig. 6: Select Commerce Companies: Commerce Patent Applications Filed v. S&P 500
(3-Months Running Average)
Fig. 6 again shows a high degree of correlation between the total patenting rate and the compounded stock prices for multiple companies. This further validates that at a broader industry level, volatility in individual stock prices and patent filings is filtered out, and the general rate of patenting is correlated with the broader Stock Market.
The correlation shown in Fig. 6 is particularly remarkable because we compared the number of patent applications filed by a limited set of 26 companies against the full S&P 500 index. This means that the patent filing activities of a representative set of companies can be correlated with the broader economic activity of a wider market, which should again be interesting for our friends in the public capital markets investment community.
Conclusions: Patent Filings Rates (and implicitly R&D Investments ) of Technology Companies Appear Correlated with Stock Market Prices
We compared the rate of patent filing for a number of companies active in the Commerce space against both their respective stock prices and against broader Stock Market indexes. As discussed above in connection with Figs. 1-6, we identified systematic correlations, suggesting that the patent filing rates, and therefore the R&D investment levels, are indeed correlated with the performance of the Stock Market.
Causality: Can Patent Filings Rates Predict Stock Performance?
Given the correlation between patent filing rates and stock prices, we further sought to test our additional hypothesis: can the patent filing rate of a public company be used to predict changes in its stock price?
Answering this question means addressing a classic dilemma of causality in the statistics of dependent random variables and stochastic processes: given demonstrable correlation between patent filing rates and stock prices, are the patent filing rates indicative of future performance of the Stock Market? Or is the variation in the patent filing rates itself caused by the fluctuation in the stock price of the respective companies? Or is there a simple non-causal correlation between the two signals when considered on a larger temporal scale, perhaps with both signals responding to a yet unidentified external driving function?
For this analysis, we used the dataset from Fig. 6, which consisted of (1) the total Commerce-related patent applications filed globally by the following 26 companies: Alibaba, Amazon, American Express, Apple, Bank of America Merrill Lynch (BOAML), Cisco, Citibank, eBay, Facebook, Google, HP, IBM, Intel, JP Morgan Chase, MasterCard, Microsoft, Oracle, PayPal, SAP, Sony, Square, Target, Visa, Walmart, Wells Fargo and Yahoo, and (2) the S&P 500 index.
We treated the compounded stock prices and the S&P 500 index prices as discrete-time stochastic processes, which we assumed to be dependent based on the analysis from Fig. 6 above. We then computed the Pearson Correlation Coefficient (Pearson CC) for the two series, which is a metric of linear dependence between random variables. This time we considered the raw data at a monthly-level, with no averages. The 3-month running averages considered in Figs. 1-5 helped remove the short-term volatility in the monthly data, but such averaging would decrease the accuracy of a more rigorous mathematical analysis.
To investigate various potential causal dependence relationships, we offset each series by 1, 2, 3 and 4 months, both forward and back in time, and we recomputed the Pearson CC each time. Fig. 7 below shows the results of this analysis:
Fig. 7: Pearson Correlation Coefficient for Select Commerce Companies: Commerce Patent Applications Filed v. S&P 500
Fig. 7 shows three correlation points in time that deserve analysis: (1) the period of 0-60 days before applications are filed compared to the S&P 500 index, (2) a lag of 4 months for the applications filed v. the S&P 500 index, and (3) a period of 60-90 days after applications are filed.
To investigate this correlation with high temporal resolution, it is important to understand how Yahoo Finance serves data via its API for monthly intervals: we retrieved the monthly data used in Fig. 7 using a Python datamining library (pandas_datareader), and Yahoo reported the figures used in Fig. 7 as being effective as of the first trading day of each month. In parallel, however, we also retrieved the full S&P 500 dataset for the 2001-2014 period on a daily basis, and we observed that the figures reported by Yahoo were actually the index price for the last day of each month. So the monthly dates shows by Yahoo were just a convenience index, and for purposes of our analysis we have to remember that we are comparing (a) total patent applications filed during each calendar month with (b) the stock price that was effective on the last day of trading during each month.
0-60 Days Lag for Applications Filed v. S&P 500
For this period we are exploring the maximum correlation signal, which is mapping patent applications during a calendar month with the S&P 500 during the same month (stronger) and the previous month (a bit weaker). As an example, given how we extracted this data from Yahoo Finance, this means that we are comparing the total number of patent applications filed during December 2014 with the S&P 500 index on December 31, 2014 (last trading day in December 2014) and November 28, 2014 (last trading day in November 2014).
The correlation shown in Fig. 7 for this period would initially suggest that the S&P 500 is the leading causal signal, and it is driving the patenting rate and the R&D investment levels.
It is important, however, to understand that for most technology companies, R&D spending is allocated at least 1-2 months in advance, and sometimes more than one quarter in advance. Analogously, after outside counsel has been instructed to file a patent application and the budget has been authorized, patent applications go through a typical logistical pre-filing process for a duration of 1-4 weeks before being filed, which often includes inventor declarations being collected from inventors who may not be located in the same geography, final adjustments to the claims and drawings, final revisions to the patent specification (e.g., writing the Abstract and Summary sections), and so on. Also, outside counsel usually receives instructions and budget authorization to file patent applications at least 30-days in advance for US applications, and often 30-60 days in advance for foreign applications. Once patent applications are authorized for filing and all the work has been done, it is highly unusual for companies to rescind the order at the last minute. Consequently, patents filed during a month are usually the result of patent filing decisions that were made during the prior 1-2 calendar months.
Revisiting the 30-60 time lag correlation shown in Fig. 7, we can now conclude with a high degree of confidence that the stock price at the end of a calendar month for a company will likely trail and be correlated with the patent filing decisions made by that company around the end of the previous calendar month, and possibly closer to 60 days in advance when considering non-US patent applications.
So for our friends investing in the public markets who had the patience to read this far, a survey of changes in patent application filing plans of a company around the end of a calendar month may serve as a good indicator of how the stock price for that company may perform 30 days later. But before you trade on that information, we should chat about insider trading considerations, or please check with your favorite Corporate attorney about SEC trading regulations.
Analogously, our friends at law firms handling large patent portfolios for US public companies may be sitting on datasets with interesting applications to public market trading. We would advise you not to trade on that data given ethical, confidentiality and other attorney-client obligations. Also watch out for old friends reaching out to you from Hedge Funds and other public investment financial institutions.
The somewhat weaker 2-moth lag Pearson Correlation Coefficient signal from Fig. 7 also deserves some further analysis since it could indicate a parallel reverse causal relationship. This may mean that the patent filing rate during a calendar month is also influenced by stock market performance around the beginning of the previous month (the end of month, two calendar months prior), which would be reasonable.
Summarizing this analysis, we could conclude that a typical public company may follow a baseline patent filing process driven by its internal R&D and operational plans during each calendar month, and in the process, it will likely (a) adjust its patent filing rate for that month to a certain extent based on economic factors that occurred during the previous month, and (b) it will eventually file a number of patent applications during that month that will ultimately prove to be an indicator for its stock price during the following month.
It is particularly interesting that on the average, the patent application filing rate appears to be better correlated with the future stock price compared to the past stock price, which could mean that companies are reacting more to their future business outlook compared to past economic conditions. Alternatively stated, the internal business outlooks of public companies eventually reach the equity markets, and their stock prices move accordingly, but by then their patent filing rates have already been adjusted accordingly.
We defer the analysis of the other two Pearson CC correlation signals from Fig. 7 (4-momth lag and 2-3 months lead) to a future article.
Correlation between R&D Spending and the Stock Market
The analysis above suggests a meaningful correlation between the patent filing rates of public companies during a particular calendar month and their stock prices approximately 30 days in the future. Since patents are normally an indicator of R&D investment, it is reasonable to also expect a correlation between R&D investment and stock performance.
In our experience, R&D leads patent filings by 1-12 months. Companies with strong patent strategies tend to file faster, particularly in the new First-to-File regime phased into the US since 2011, which incentivizes companies to file patents quickly to maximize priority rights. In contrast, companies that do not particularly focus on patents are more reactive and tend to file later. Filing patent applications more than 12 months after the R&D activities occurred could result in loss of patent rights both the US and abroad, so we can normally consider 12 months to be an outer window.
Assuming a mean lead time of 6 months for R&D investment relative to patent applications, the analysis in Fig. 7 could be interpreted to suggest that R&D investment could be used to forecast stock performance about two financial quarters in the future. In practice we would discourage this conclusion without further analysis, given that intervening events could alter this correlation significantly. We may analyze this correlation in a future article. But there is no doubt that some degree of correlation between R&D activities and stock performance 3-6 months in the future should exist.
Potential Areas of Further Inquiry
The Pearson Correlation Coefficient analysis discussed in connection with Fig. 7 above considered a broader set of companies together. It is possible that the conclusions reached there may be different or may need to be adjusted for individual companies, particularly companies that have diversified operations, are patenting at a rate that is decorrelated from their R&D investment spending, or are experiencing artificially inflated or deflated valuations. Even in those cases, this analysis may still hold for shorter time intervals.
Here are some areas of analysis that could be considered to validate and further explore the correlation and causality relationships discussed above and increase the level of confidence in these conclusions:
Explore this correlation sensitivity for companies that have lower or higher degrees of concentration in Commerce (see our analysis here).
The Pearson Correlation Coefficient is an indication of linear correlation, but does not capture nonlinear or more complex aspects of correlation.Expand the statistical analysis to improve the sophistication and accuracy of the correlation analysis.
Consider a broader set of patents, particularly for companies with a higher R&D concentration outside Commerce;
Explore the correlation sensitivity to US-only patent application filings v. global patent portfolios.
Consider broader time ranges, including patent applications filed more recently.Investigate how the correlation results may change for more recent patent filings.
The variation of the Pearson Correlation Coefficient in Fig. 7 is relatively low in absolute terms.Validate that the variations shown here are indeed meaningful.
Attempt to improve the prediction window discussed above using a better selection of patents and a more carefully selected set of companies.
Run a historical analysis of retroactive predictions and compare against actual historical market performance to confirm that in practice these conclusions hold.
Explore practical and legal mechanisms for collecting early patent filing information that could be actionable to predict future stock market performance.
Explore the ability use statistically-relevant public patent filing information to infer the actual patent filing rates in real time, which would obviate the need to collect information from private sources.
A Word of Caution about Trading
We feel that it is critical to emphasize to anyone who would actually consider trading stocks based on our analysis that (a) trading on private information may be illegal in many cases, both in the US and abroad, so you are strongly advised to consult with competent counsel, and (b) this analysis was based on certain assumptions that may be incorrect in whole or in part, and our conclusions and results may not hold in practice either in totality or in specific applications, so more analysis must be done to validate the analysis framework and the results before acting in any way on this information.
About this Analysis
For this analysis, we identified a set of over 1 million patents and applications for the period 1995 – 2016 that cover a wide span of Commerce technologies and business models, and we extracted the patents and applications for 26 prominent public companies listed on the New York Stock Exchange or NASDAQ. We limited the timeframe of the analysis to the period between January 2001 and December 2014 and we considered global patent filings.
This analysis was limited to patents in the Commerce space, including the following:
Payment space and Financial Technology (FinTech):payments, payment processing, credit processing, traditional currencies and crypto currencies, credit card and ATM card processing, ATM and credit banking systems, SWIFT transactions and security technology, block chain technology (including distributed ledger and consensus-based algorithms), payment gateways, Point of Sale technology, credit card readers and other credit card processing equipment, and related data processing and communications technologies.
Omnichannel Technologies: digital coupons, digital offers, commerce data collection and data analysis, SKU-level data collection and processing, Consumer Relationship Management (CRM) systems, Enterprise Resource Planning (ERP) systems, cloud-based platforms, desktop computers and other devices, electronic tablets (iPad, Android, Windows and other operating systems and form factors), Application Programming Interfaces (APIs), data analytics, data monetization, table reservations, order deliveries, order management, mobile ordering using phones and tablet devices, and other forms of business digital transformation.
The data sets were too large to be processed via normal patent searches and Excel spreadsheets, so most of the analytical work was conducted using Python and the final results were synthesized in Excel. We also used Python to automatically retrieve stock prices for the companies of interest and for market indexes through the Yahoo Finance API.
If you would like to discuss any aspect of this article please do not hesitate to contact us at email@example.com.
Note: This analysis was performed programmatically, without a review of actual claims. The opinions in this article are limited to the scope of this article, and do not necessarily reflect our opinions in general, or the opinions of any of our clients or partners.