WASHINGTON Jan 15 (Reuters) - Three academics say they have come up with a way for Wall Street regulators to harness powerful computer algorithms to gauge how well public feedback is received and incorporated into the rules they write.
The algorithm, dubbed "RegRank," was unveiled late last week in a new paper authored by the Commodity Futures Trading Commission's former Chief Economist Andrei Kirilenko, University of Maryland professor Shawn Mankad, and University of Michigan professor George Michailidis.
The algorithms and statistical tests at the heart of the paper are fairly standard. But what the researchers say is unique is how they use these empirical tools to measure "regulatory sentiment" and "test the impact" of the public comment process on rulemaking.
Such a tool could come in handy for regulators such as the CFTC and Securities and Exchange Commission, both of whom have faced legal challenges to rules on the grounds that regulators failed to properly incorporate feedback, as required by law.
"The regulators need to be accountable to the public," Kirilenko said of the rulemaking process.
It could also be helpful for the industry and the public, as a way to ensure that regulators heed feedback about flaws in proposed reforms. The paper can be found at
Kirilenko, now a professor at the Massachusetts Institute of Technology's business school, told Reuters in an interview the idea stems from his time as CFTC chief economist.
In that role, Kirilenko was in charge of reviewing cost-benefit analyses for new rules for over-the-counter derivatives and helping to ensure public comments - drafted by lawyers, lobbyists and industry insiders - were weighed.
During that process, he realized all the legal jargon often used in the comments could actually be boiled down to clusters of "basic words," he said.
"I thought this is an interesting operation," he said. "Let's see if we can actually uncover this hidden structure using a computer algorithm."
The technology is an automated "machine-learning" method that works by mining regulatory text, which can span hundreds of pages. The algo searches for certain clusters of key words that are deemed "pro-regulation" or "anti-regulation."
To measure whether words are positive or negative, the researchers combined two publicly available dictionaries created by computer scientists that label words with a positive or negative tone.
Positive words could include, for instance, "bullish," "affordable" or "advantageous," while negative words include "exacerbate" or "corrupt."
The algo is then able to keep score of how often such pro or anti-regulatory words appear, allowing researchers to compare how a proposed rule evolves into a final one and whether the public's comments were heeded.
The topic has gained more attention as regulators have drafted hundreds of new rules required by the 2010 Dodd-Frank Wall Street reform law.
The Administrative Procedures Act (APA) requires many federal agencies to carefully weigh public feedback, provide notice and comment, and afford due process in general before rules are finalized.
The CFTC is currently facing a legal challenge by three Wall Street trade groups over its policy governing how U.S. swaps rules should apply overseas.
The groups allege the agency violated the APA, in part by failing to seek proper public comments and weigh costs and benefits.
As a testing ground for the "RegRank" algo, the three researchers said in their Jan. 11 paper that they examined 104 proposed and 67 final rules at the CFTC, as well as 60,000 public comments submitted between January 2010 and September 2013, as the agency was busy drafting new rules for the over-the-counter derivatives market required by the Dodd-Frank law.
The researchers determined that "the government agency does adjust the final rule in response to public comments."
The paper still must be peer-reviewed, and the researchers plan to present it at conferences.
"The results show that the government listens, but more research would be needed for deeper insights," said Mankad.