The latest articles on DEV Community by Daniel Oliveira (@daniel_oliveira_4bd55d35d). https://dev.to/daniel_oliveira_4bd55d35d https://media2.dev.to/dynamic/image/width=90,height=90,fit=cover,gravity=auto,format=auto/https:%2F%2Fdev-to-uploads.s3.us-east-2.amazonaws.com%2Fuploads%2Fuser%2Fprofile_image%2F4071986%2Fe3ad517f-a098-4b70-bf03-148835a4f133.jpg https://dev.to/daniel_oliveira_4bd55d35d en Daniel Oliveira Fri, 06 Nov 2026 00:45:48 +0000 https://dev.to/daniel_oliveira_4bd55d35d/when-does-applied-materials-amat-report-q4-2026-earnings-and-what-do-analysts-expect-4goa https://dev.to/daniel_oliveira_4bd55d35d/when-does-applied-materials-amat-report-q4-2026-earnings-and-what-do-analysts-expect-4goa <p><a href="https://equibles.com/stocks/amat/summary" rel="noopener noreferrer">Applied Materials</a> reports its fiscal fourth quarter of 2026 next week, with the stock 56% below its June record after a 13.6% drop on November 5. These are the questions people ask about it, each answered with one number and the page where the data lives. At the end I list the API calls behind each answer.</p> <h2> When does Applied Materials report Q4 2026 earnings? </h2> <p>Thursday, November 12, 2026, after the market closes, with the call at 1:30 p.m. Pacific (21:30 UTC). The call page will carry the transcript: <a href="https://equibles.com/stocks/amat/calls/2026-q4" rel="noopener noreferrer">Applied Materials Q4 2026 earnings call</a>. Earlier calls are on <a href="https://equibles.com/stocks/amat/earnings-calls" rel="noopener noreferrer">its earnings calls page</a>.</p> <h2> What did Applied Materials guide? </h2> <p>Revenue of $10.25 billion, plus or minus $500 million, and non-GAAP earnings of $4.02 a share, plus or minus $0.20, in its third-quarter release. That would be 12.5% more than the record $9.12 billion of the third quarter: <a href="https://equibles.com/stocks/amat/documents/d1dc8ec0-db98-4799-8485-0065aa5c4dc7" rel="noopener noreferrer">the Q3 2026 earnings release</a>.</p> <h2> What do analysts expect? </h2> <p>$10.30 billion on average from 24 analysts, just above the guide's midpoint, with a range of $10.25 billion to $10.54 billion, and $4.06 a share. Figures as of November 1: <a href="https://equibles.com/stocks/amat/analyst-estimates" rel="noopener noreferrer">Applied Materials analyst estimates</a>.</p> <h2> Does Applied Materials usually beat its guide? </h2> <p>By a little. Revenue came in 1.8% to 3.4% above the midpoint of its own range in each of the last three quarters: <a href="https://equibles.com/stocks/amat/financials" rel="noopener noreferrer">Applied Materials financials</a>.</p> <h2> How has the stock reacted to past reports? </h2> <p>Badly, even after beats. It fell the session after six of its last eight reports, including -14.1% in August 2025 and -5.1% in August 2026, and moved 6.5% on average either way: <a href="https://equibles.com/stocks/amat/summary" rel="noopener noreferrer">Applied Materials stock page</a>.</p> <h2> How much of Applied Materials is sold short? </h2> <p>14.2 million shares at the October 15 settlement, 0.6 million more than two weeks earlier, or 2.2 days of average volume: <a href="https://equibles.com/stocks/amat/short-interest" rel="noopener noreferrer">Applied Materials short interest</a>.</p> <h2> Have insiders been selling? </h2> <p>Yes: 42 open-market sales worth about $179 million this year, the latest on October 2, and no open-market purchases: <a href="https://equibles.com/stocks/amat/insider-trading" rel="noopener noreferrer">Applied Materials insider trading</a>.</p> <h2> Where is the stock going into the report? </h2> <p>It closed at $319.25 on November 5, down 13.6% that day as chip-equipment stocks fell together (Lam Research -16.4%, KLA -13.4%). That is 55.8% below its record close of $723.00 on June 30, a market value of about $253 billion.</p> <h2> How I pulled these numbers </h2> <p>Each answer is one call to the <a href="https://equibles.com/mcp" rel="noopener noreferrer">Equibles MCP server</a>: <code>GetUpcomingInvestorEvents</code> for the call time, <code>GetAnalystEstimates</code> for consensus, <code>GetGuidance</code> and <code>ReadDocumentLines</code> on the 8-K releases for the guidance history, <code>GetStockPrices</code> for the post-report moves, <code>GetShortInterest</code> for the FINRA settlements and <code>GetInsiderTransactions</code> for the Form 4s.</p> <p>Data: SEC filings, company releases, analyst estimates, FINRA short interest and daily prices, processed by Equibles.</p> <p>This is our analysis of public filings and market data, not investment advice. Equibles staff may hold positions in the securities discussed.</p> api Daniel Oliveira Wed, 04 Nov 2026 15:03:43 +0000 https://dev.to/daniel_oliveira_4bd55d35d/parsing-ipo-lock-ups-from-440-prospectuses-the-90-days-that-arent-and-other-traps-1gk7 https://dev.to/daniel_oliveira_4bd55d35d/parsing-ipo-lock-ups-from-440-prospectuses-the-90-days-that-arent-and-other-traps-1gk7 <p>Every US IPO prospectus describes its lock-up: who agreed not to sell, for how long, and what can end it early. No structured feed carries those terms, so I read them out of the text for 440 IPOs from January 2024 to October 2026. The headline numbers came out clean (89% use 180 days or six months, 26 deals write in an early release, and on the Nasdaq Capital Market 154 of 180 lock-ups bind only insiders and the largest holders), but my first pass got a lot of them wrong. These are the traps, roughly in the order I hit them.</p> <h2> 1. Finding the IPOs is its own problem </h2> <p>EDGAR full-text search is the easiest way in. I searched final prospectuses (Form 424B4) for the phrase "there has been no public market", which nearly every IPO prospectus uses, plus "lock-up". That returned 824 filings. Then:</p> <ul> <li> <strong>Blank-check companies.</strong> 314 were SPACs (SIC code 6770). Their lock-ups work differently, so they go.</li> <li> <strong>Direct listings and resale registrations.</strong> These also say there has been no public market, but they register existing holders' shares, and their "lock-up" passages usually say that nobody has one. 25 of them.</li> <li> <strong>Duplicates.</strong> A repriced deal files a second 424B4 days later. Keep the later one if it follows within 30 days, otherwise keep the first: a later 424B4 can be a follow-on offering that repeats the phrase.</li> </ul> <h2> 2. The text fights back </h2> <p>Prospectuses are HTML generated by filing agents, and the same words come out in many shapes:</p> <ul> <li> <code>180&amp;nbsp;days</code> and zero-width spaces between words. Normalize <code>\xa0</code> and <code>​</code> to spaces before any regex.</li> <li> <code>lock -up</code>, <code>lock- up</code>, <code>lockup</code> and <code>Lock-Up</code>, often in the same document.</li> <li>Periods written as <code>180 days</code>, <code>180-day</code>, <code>one hundred eighty (180) days</code>, <code>six (6) months</code>, <code>6 months</code> and <code>until the 181st day</code>.</li> </ul> <p>Sentence splitting breaks too. "J.P. Morgan Securities LLC" and "Morgan Stanley &amp; Co. LLC" each contain periods, and the lock-up sentence usually names the underwriters right before the parties. Splitting on <code>.</code> cut the subject off half my sentences, so I switched to a fixed window of characters before each match.</p> <h2> 3. The 90 days that aren't the lock-up </h2> <p>My first parser took the most common period mentioned near "lock-up". In small deals it kept answering 90 days. The prospectuses describe two restrictions side by side: the company agrees not to issue new shares for 90 days or three months, and the directors, officers and big holders agree not to sell for six months. The company's promise is often repeated more times, so it wins a count.</p> <p>The fix was to keep only periods whose preceding text names holders, and to prefer a long one when both appear:<br> </p> <div class="highlight js-code-highlight"> <pre class="highlight python"><code><span class="n">HOLDER</span> <span class="o">=</span> <span class="n">re</span><span class="p">.</span><span class="nf">compile</span><span class="p">(</span><span class="sa">r</span><span class="sh">'</span><span class="s">director|officer|shareholder|stockholder|holders? of|securityholder</span><span class="sh">'</span><span class="p">,</span> <span class="n">re</span><span class="p">.</span><span class="n">I</span><span class="p">)</span> <span class="k">def</span> <span class="nf">insider_period</span><span class="p">(</span><span class="n">passages</span><span class="p">):</span> <span class="n">found</span> <span class="o">=</span> <span class="nc">Counter</span><span class="p">()</span> <span class="k">for</span> <span class="n">text</span> <span class="ow">in</span> <span class="n">passages</span><span class="p">:</span> <span class="k">for</span> <span class="n">pos</span><span class="p">,</span> <span class="n">days</span> <span class="ow">in</span> <span class="nf">periods_in</span><span class="p">(</span><span class="n">text</span><span class="p">):</span> <span class="c1"># (offset, length in days) </span> <span class="k">if</span> <span class="n">HOLDER</span><span class="p">.</span><span class="nf">search</span><span class="p">(</span><span class="n">text</span><span class="p">[</span><span class="nf">max</span><span class="p">(</span><span class="mi">0</span><span class="p">,</span> <span class="n">pos</span> <span class="o">-</span> <span class="mi">700</span><span class="p">):</span><span class="n">pos</span><span class="p">]):</span> <span class="n">found</span><span class="p">[</span><span class="n">days</span><span class="p">]</span> <span class="o">+=</span> <span class="mi">1</span> <span class="n">long_</span> <span class="o">=</span> <span class="nc">Counter</span><span class="p">({</span><span class="n">d</span><span class="p">:</span> <span class="n">n</span> <span class="k">for</span> <span class="n">d</span><span class="p">,</span> <span class="n">n</span> <span class="ow">in</span> <span class="n">found</span><span class="p">.</span><span class="nf">items</span><span class="p">()</span> <span class="k">if</span> <span class="n">d</span> <span class="o">&gt;=</span> <span class="mi">120</span><span class="p">})</span> <span class="nf">return </span><span class="p">(</span><span class="n">long_</span> <span class="ow">or</span> <span class="n">found</span><span class="p">).</span><span class="nf">most_common</span><span class="p">(</span><span class="mi">1</span><span class="p">)[</span><span class="mi">0</span><span class="p">][</span><span class="mi">0</span><span class="p">]</span> <span class="k">if</span> <span class="n">found</span> <span class="k">else</span> <span class="bp">None</span> </code></pre> </div> <p>That moved most of the false 90-day lock-ups to six months. A hand check of 30 random prospectuses then found the right period in 29.</p> <h2> 4. "Earlier of" is where the real date hides </h2> <p>An early-release clause doesn't announce itself. The common forms:</p> <ul> <li> <strong>Earnings-based:</strong> the period ends on "the earlier of (i) the second trading day after we publicly announce earnings for the second quarter following the most recent period for which financial statements are included in this prospectus and (ii) 180 days". To date it you need the company's earnings 8-Ks (Item 2.02) and the latest period in the prospectus.</li> <li> <strong>Price-based:</strong> a slice is released if the stock closes 25% or 30% above the IPO price on five of ten trading days.</li> <li> <strong>Staged:</strong> tranches on fixed days. SpaceX's prospectus lists 15 release dates, from two full trading days after its first earnings report to mid-2027.</li> <li> <strong>"Later of":</strong> Chime's employee release starts at the later of an earnings date and day 90, so a pattern that looks only for "earlier of" misses it.</li> </ul> <p>One false positive is everywhere: the blackout extension. Many large-company lock-ups say that if day 180 falls inside a trading blackout, the period runs a few days longer. That is the opposite of an early release, and it uses the same words ("earnings", "trading day"). I dropped any earnings match whose sentence mentions a blackout, then read all 26 early-release deals by hand.</p> <h2> 5. Who signed is a different question from how long </h2> <p>A lock-up only binds the people who sign it. On the larger exchanges the parties are usually "our directors, executive officers and the holders of substantially all of our outstanding securities". On the Capital Market they're usually "our directors, officers and holders of 5% or more".</p> <p>Classifying that sentence had its own traps:</p> <ul> <li> <strong>The wrong sentence.</strong> "Shares purchased by our directors or officers in the directed share program will be subject to lock-up" matches "directors" and "lock-up" but says nothing about who signed. Skip sentences about directed share programs, Rule 701 and 10b5-1 plans.</li> <li> <strong>"Certain" changes everything.</strong> "Holders of the outstanding ordinary shares" means everyone; "certain holders of the outstanding ordinary shares" doesn't.</li> <li> <strong>Thresholds vary.</strong> 5% is the most common, but I saw 1%, 2% and 10%, and one deal that locked up "holders of not less than 95%".</li> </ul> <p>A hand check of 20 found 19 right. The classifier still reads wording, not share counts: in a founder-owned company, "directors, officers and 5% holders" can cover nearly every share.</p> <h2> 6. Which date is day 0? </h2> <p>Lock-ups run from the prospectus date, and the 424B4 is filed a day or two after it. Using the filing date minus one business day was wrong for about half the deals, usually by a day, so I read the stated date: "The date of this prospectus is June 3, 2026", or on many covers just the date printed after the underwriters' names, right before the table of contents. Some covers print it in capitals, and a few stated dates turned out to be a different document's, so I kept a stated date only if it fell within five days before the filing; Rule 424(b) requires the filing within two business days.</p> <p>Then the counting convention. "For a period of 180 days after the date of this prospectus" restricts sales through day 180. SpaceX's own table lists day 180 as the earliest date the last tranche is available. Rather than pick one reading, the event test uses a three-day window from the day before to the day after the first sale day, which contains it either way.</p> <h2> What came out of it </h2> <p>On the day a lock-up ended, the typical stock didn't fall: across 330 standard lock-ups, the median three-day return against SPY was +0.7%, against -1.2% for the same stocks a month earlier. Volume did jump, a median 1.34 times its prior level for big deals where substantially all holders signed, and didn't move on the Capital Market, where most lock-ups name only the insiders. The full study, with the charts, an expiry calendar for November to January, and the method:</p> <p><a href="https://equibles.com/research/who-is-actually-locked-up-after-an-ipo" rel="noopener noreferrer">https://equibles.com/research/who-is-actually-locked-up-after-an-ipo</a></p> <p>Has anyone found a way to get lock-up waivers systematically? Underwriters can release shares at any time, and only some deals announce it.</p> <p>Data: SEC prospectuses and daily prices, processed by Equibles.</p> analysis data webscraping Daniel Oliveira Wed, 04 Nov 2026 06:40:02 +0000 https://dev.to/daniel_oliveira_4bd55d35d/a-one-sentence-stock-screen-in-claude-and-three-ways-it-can-quietly-mislead-you-1ao4 https://dev.to/daniel_oliveira_4bd55d35d/a-one-sentence-stock-screen-in-claude-and-three-ways-it-can-quietly-mislead-you-1ao4 <p>The Equibles MCP server has a <code>ScreenStocks</code> tool, so with it connected to Claude or ChatGPT a one-sentence description of the stocks you want becomes a set of filters and a table. I ran one screen through it on November 3, 2026, and it showed three ways a screen can quietly mislead you. Every figure below is from those tool calls.</p> <h2> The sentence and the filters </h2> <blockquote> <p>Find US stocks worth more than $2 billion that are profitable, grew revenue more than 20% in their latest quarter, have gross margins above 50% and trade below 30 times earnings. Skip anything that trades less than $10 million a day.</p> </blockquote> <p>That sentence maps to six arguments:<br> </p> <div class="highlight js-code-highlight"> <pre class="highlight json"><code><span class="p">{</span><span class="w"> </span><span class="nl">"minMarketCap"</span><span class="p">:</span><span class="w"> </span><span class="mi">2000000000</span><span class="p">,</span><span class="w"> </span><span class="nl">"minNetIncome"</span><span class="p">:</span><span class="w"> </span><span class="mi">0</span><span class="p">,</span><span class="w"> </span><span class="nl">"minRevenueGrowth"</span><span class="p">:</span><span class="w"> </span><span class="mi">20</span><span class="p">,</span><span class="w"> </span><span class="nl">"minGrossMargin"</span><span class="p">:</span><span class="w"> </span><span class="mi">50</span><span class="p">,</span><span class="w"> </span><span class="nl">"maxPe"</span><span class="p">:</span><span class="w"> </span><span class="mi">30</span><span class="p">,</span><span class="w"> </span><span class="nl">"minDollarVolume"</span><span class="p">:</span><span class="w"> </span><span class="mi">10000000</span><span class="w"> </span><span class="p">}</span><span class="w"> </span></code></pre> </div> <p>32 stocks matched. The reply also dated every column: prices from the November 2 close, short interest from FINRA's October 15 settlement, and 13F holder counts from the quarter ended June 30. That's the first habit worth keeping: ask the assistant to show its arguments, and check that every number in your sentence became exactly one filter.</p> <h2> Trap 1: a missing number removes the stock </h2> <p>I then ran the same screen on a watchlist (the tool takes a <code>tickers</code> list): Apple, Microsoft, Amazon, Reddit, Palantir, SoFi, Robinhood and Uber. Only Reddit passed, which I expected. What I didn't expect was that Robinhood and Uber didn't fail any filter. They vanished, because the screen excludes a stock missing any bounded metric, and neither has a gross margin on file.</p> <p>If you build or use a screener, "no data" and "fails the test" should look different. Ask the assistant which filter removed each stock.</p> <h2> Trap 2: the same growth number can mean different things </h2> <p>Sorted by revenue growth, the top ten were Micron (379%), Sandisk (378%), Royal Gold (115%), Nvidia (106%), Chord Energy (84%), TG Therapeutics (70%), Franco-Nevada (57%), Western Digital (56%), Reddit (53%) and Hinge Health (53%). Memory makers, gold royalty companies and an oil producer sell commodities, so the prices of what they sell move their revenue. That's a different kind of growth from Reddit's or Hinge Health's, and the filter can't tell them apart.</p> <h2> Trap 3: cheap is a question, not an answer </h2> <p>Reddit passed at 18.7 times trailing earnings. <code>GetValuationMultiplesHistory</code> showed it at 53.4 times in early February. Earnings rose over that time (Q3 net income $283.3M against $162.7M a year earlier), so the multiple fell because the price fell: $129.60 on October 29, $91.73 on October 30, down 29.2% the day after the report.</p> <p><code>GetEarningsBrief</code> gave the reason in the CEO's own words: search referrals fell by more than a million a week in October, and if that continued, daily active uniques "could fall in Q4 by a low double-digit percentage." A screen finds what's cheap; one more prompt tells you why.</p> <h2> The one-prompt version </h2> <blockquote> <p>Screen US stocks worth more than $2 billion that are profitable, grew revenue more than 20% last quarter, have gross margins above 50%, trade below 30 times earnings and over $10 million a day. Sort by revenue growth. For the three most familiar names, compare the valuation with peers and with the company's own history, explain any big price drop since the last report from the earnings call, and list insider sales and short interest. Give the date of every number.</p> </blockquote> <p>The full walkthrough, with six prompts, every answer and how to check each one, is here:</p> <p><a href="https://equibles.com/blog/how-to-screen-for-stocks-with-chatgpt-or-claude" rel="noopener noreferrer">https://equibles.com/blog/how-to-screen-for-stocks-with-chatgpt-or-claude</a></p> <p>How do you handle missing data in your own screens: exclude, flag, or fill?</p> <p>Data: SEC filings, FINRA short interest, 13F filings and market prices, processed by Equibles.</p> ai claude fintech mcp Daniel Oliveira Wed, 04 Nov 2026 00:17:42 +0000 https://dev.to/daniel_oliveira_4bd55d35d/when-does-archer-aviation-achr-report-q3-2026-earnings-and-what-do-analysts-expect-50mp https://dev.to/daniel_oliveira_4bd55d35d/when-does-archer-aviation-achr-report-q3-2026-earnings-and-what-do-analysts-expect-50mp <p><a href="https://equibles.com/stocks/achr/summary" rel="noopener noreferrer">Archer Aviation</a> reports its third quarter of 2026 next week. These are the questions people ask about it, each answered with one number and the page where the data lives. At the end I list the API calls behind each answer.</p> <h2> When does Archer report Q3 2026 earnings? </h2> <p>Monday, November 9, 2026, after the market closes, with the webcast at 2:00 p.m. Pacific (22:00 UTC). The call page will carry the transcript: <a href="https://equibles.com/stocks/achr/calls/2026-q3" rel="noopener noreferrer">Archer Q3 2026 earnings call</a>. Earlier calls are on <a href="https://equibles.com/stocks/achr/earnings-calls" rel="noopener noreferrer">Archer's earnings calls page</a>.</p> <h2> What revenue do analysts expect? </h2> <p>$3.57 million on average from 7 analysts, with a range of $2.0 million to $6.0 million. Two analysts publish an EPS estimate, averaging a loss of $0.245 a share. Figures as of October 14: <a href="https://equibles.com/stocks/achr/analyst-estimates" rel="noopener noreferrer">Archer analyst estimates</a>.</p> <h2> What do we forecast? </h2> <p>$5.3 million, above the consensus, because the second quarter already brought in $5.0 million with a full quarter of the Hawthorne airport fueling business, in which Archer bought a 75% stake on April 1. All three revenue lines (fueling and ground services, leases, other) sit at Hawthorne. The model is in our <a href="https://equibles.com/research/archer-aviation-achr-q3-2026-earnings-preview-and-forecast" rel="noopener noreferrer">Archer Q3 2026 earnings preview</a>.</p> <h2> How much cash is Archer burning? </h2> <p>The company guides to an adjusted EBITDA loss of $170 million to $200 million for the quarter; we put Q3 at $185 million. That takes cash and short-term investments to about $1.36 billion by our estimate, from $1.56 billion at the end of June: <a href="https://equibles.com/stocks/achr/financials" rel="noopener noreferrer">Archer financials</a>.</p> <h2> What does the Boeing deal change? </h2> <p>Boeing gets about 152 million Archer shares at closing, 19.75% of today's share count, plus two warrants worth $100 million each at $13.00 and $17.88. One of the businesses, Insitu, brings more than $200 million of revenue a year by Archer's own estimate, which is why the 2027 consensus runs from $35 million to $325.5 million: some analysts include the deal and some don't. Details in the <a href="https://equibles.com/research/archer-aviation-achr-q3-2026-earnings-preview-and-forecast" rel="noopener noreferrer">preview</a>.</p> <h2> How much of Archer is sold short? </h2> <p>114.8 million shares at the October 15 settlement, 14.9% of the 770 million shares outstanding, up 22% in a month and 4.3 days of average volume. FINRA publishes the October 30 figure on November 10: <a href="https://equibles.com/stocks/achr/short-interest" rel="noopener noreferrer">Archer short interest</a>.</p> <h2> How big a move do options price in? </h2> <p>About 9% either way by November 13, the first expiry after the report, at the November 2 close of $4.17: <a href="https://equibles.com/stocks/achr/options-chain" rel="noopener noreferrer">Archer options chain</a>.</p> <h2> Have insiders been buying? </h2> <p>Not recently. The last open-market purchases were in late 2024: Stellantis bought $5.0 million of stock at $6.65 on December 11, and CEO Adam Goldstein bought 19,762 shares at $5.12 on November 21: <a href="https://equibles.com/stocks/achr/insider-trading" rel="noopener noreferrer">Archer insider trading</a>.</p> <h2> How I pulled these numbers </h2> <p>Each answer is one call to the <a href="https://equibles.com/mcp" rel="noopener noreferrer">Equibles MCP server</a>, which I work on: <code>GetUpcomingInvestorEvents</code> for the call time, <code>GetAnalystEstimates</code> for consensus, <code>GetShortInterest</code> for the FINRA settlements, <code>GetOptionChain</code> for the straddle and <code>GetInsiderTransactions</code> filtered to open-market buys. The forecast and the Boeing math come from the 10-Q and the deal filings, in the preview.</p> <p>Data: SEC filings, analyst estimates, FINRA short interest and option prices, processed by Equibles.</p> <p>This is our analysis of public filings and market data, not investment advice. Equibles staff may hold positions in the securities discussed.</p> finance investing ai mcp Daniel Oliveira Tue, 03 Nov 2026 14:53:49 +0000 https://dev.to/daniel_oliveira_4bd55d35d/i-used-claude-with-an-mcp-server-to-prep-for-an-earnings-report-six-prompts-and-the-tool-calls-4io2 https://dev.to/daniel_oliveira_4bd55d35d/i-used-claude-with-an-mcp-server-to-prep-for-an-earnings-report-six-prompts-and-the-tool-calls-4io2 <p>Before a company reports, I want six things: the date, the consensus, management's guidance, last quarter's call, the options-implied move and the short interest. They live in six different places. Lately I've been asking for them in Claude with the Equibles MCP server connected, and I wanted to write down what the assistant actually calls behind each prompt.</p> <p>The example is SoundHound (SOUN), which reports Q3 on November 5. Every answer below came from tool calls on November 3.</p> <h2> The six prompts </h2> <ol> <li> <strong>"When does SoundHound report Q3, and when is the call?"</strong> The assistant calls <code>GetUpcomingInvestorEvents</code>: November 5, 22:00 UTC (5 p.m. Eastern), with a link to the event on SoundHound's investor site.</li> <li> <strong>"What do analysts expect, and has it moved?"</strong> <code>GetAnalystEstimates</code>: revenue of $62.18M on average from 6 analysts, EPS of -$0.05 from 5, in a snapshot captured October 29.</li> <li> <strong>"What's the 2026 revenue guidance, what does it leave for the second half, and has anything happened since that changes it?"</strong> <code>GetGuidance</code>, <code>GetFinancialStatement</code> and <code>GetInvestorRelationsNews</code>: $230M to $260M, set August 5; first-half revenue of $106.1M, so the second half needs $123.9M to $153.9M; and a LivePerson acquisition that closed September 4.</li> <li> <strong>"Summarize the last earnings call."</strong> <code>GetEarningsBrief</code>: record Q2 revenue of $61.9M, up 45%, a GAAP net loss of $42.8M, and the CFO's quote raising the outlook.</li> <li> <strong>"What move are options pricing for the report?"</strong> <code>GetOptionExpirations</code>, then <code>GetOptionChain</code> for November 6: the $4 straddle cost about $0.52 at the midpoints with the stock at $4.15, a move of about 12.5%.</li> <li> <strong>"How much of the stock is sold short?"</strong> <code>GetShortInterest</code>: 170.8M shares at the October 15 settlement, 10.1 days of average volume to cover.</li> </ol> <h2> The prompt that mattered most was the third </h2> <p>My first version of prompt 3 asked only for the guidance and what it left for the second half. The answer was correct and misleading. The guidance tool's newest source was the August 5 call, and SoundHound closed its acquisition of LivePerson on September 4. LivePerson booked $52.5M of revenue in Q2 on its own, and on the August call the CEO said the outlook would be updated to include it once the deal closed.</p> <p>So reported Q3 revenue will include 27 days of LivePerson, while the $62.2M consensus is barely above SoundHound's own $61.9M in Q2. Comparing the result with that consensus without knowing about the deal would mislead you.</p> <p>Adding "has anything happened since that changes it?" made the assistant read the investor relations news and the 8-K list, and the deal showed up. That's the general lesson for any tool-using assistant: a tool answers from what's on file as of its last source, so ask for the date of every number and for what has been filed since.</p> <h2> What else I'd tell someone setting this up </h2> <ul> <li> <strong>Ask for the as-of time.</strong> The consensus was almost five days old; option quotes carry a timestamp to the second; FINRA short interest comes out about a week and a half after its settlement date.</li> <li> <strong>Check the arithmetic once by hand.</strong> 44.2 + 61.9 = 106.1, and 230 - 106.1 = 123.9. After that, let the assistant do it.</li> <li> <strong>Combine the prompts once each one works:</strong> "Build me an earnings prep sheet for SoundHound's Q3 2026 report: the date and call time, the consensus with its date, the 2026 guidance and what it leaves for the second half, anything filed since that changes it, a summary of last quarter's call, the options-implied move for the first expiry after the report, and the latest short interest. Cite the source and date of every number."</li> </ul> <p>The full walkthrough, with every answer, how to check each one and the setup steps for ChatGPT and Claude, is here:</p> <p><a href="https://equibles.com/blog/how-to-research-a-stock-before-earnings-with-chatgpt-or-claude" rel="noopener noreferrer">https://equibles.com/blog/how-to-research-a-stock-before-earnings-with-chatgpt-or-claude</a></p> <p>What do you check before a company reports that isn't on this list?</p> <p>Data: SEC filings, FINRA short interest, company investor relations pages and option quotes, processed by Equibles.</p> ai claude mcp productivity Daniel Oliveira Tue, 03 Nov 2026 00:04:22 +0000 https://dev.to/daniel_oliveira_4bd55d35d/substantial-doubt-is-easy-to-grep-and-easy-to-misread-notes-from-checking-every-10-q-for-1lo3 https://dev.to/daniel_oliveira_4bd55d35d/substantial-doubt-is-easy-to-grep-and-easy-to-misread-notes-from-checking-every-10-q-for-1lo3 <p>Every quarter, US companies have to decide whether there is "substantial doubt about an entity's ability to continue as a going concern" over the next year, and if there is, say so in their 10-Q. It's one of the few places where a company tells you in its own words that it might run out of money.</p> <p>I recently counted how many companies say it, using the latest 10-Q filed since July 1 for 4,899 companies. The headline: one in six Nasdaq operating companies (382 of 2,281) carries the warning, against 56 of 1,670 on the NYSE and NYSE American. Getting to those numbers took a lot more than a full-text search. Here's what tripped me up.</p> <h2> 1. The phrase shows up in sentences that say the opposite </h2> <p>Searching for "substantial doubt" finds the warnings, but also:</p> <ul> <li> <strong>Alleviated doubt.</strong> "These conditions raise substantial doubt... management's plans alleviate the substantial doubt." That's not a warning. 107 operating companies wrote some version of it.</li> <li> <strong>Conditional risk factors.</strong> "If we are unable to raise additional capital when needed, that could raise substantial doubt" describes a hypothetical. Damora Therapeutics, worth $1.3 billion, uses exactly that sentence while saying its cash funds operations into the second half of 2029.</li> <li> <strong>The negative.</strong> A plain sufficiency statement ("sufficient to fund operations for at least 12 months from the issuance date") or "substantial doubt does not exist". Hundreds of reports discussed going concern like this without a warning.</li> <li> <strong>Someone else's doubt.</strong> National Health Investors, a healthcare REIT, wrote that a tenant has been on a cash basis since 2022, when it "received financial information from Bickford raising substantial doubt about its ability to continue as a going concern". The REIT itself carries no warning.</li> </ul> <p>Spyre Therapeutics makes a good test case: "If we are unable to raise additional capital when needed, we will not be able to continue as a going concern." Strong words, but still conditional.</p> <p>So I sorted each report into five buckets: warning, alleviated, discussion without a warning, nothing found, and unclear. The 42 unclear reports were left out rather than guessed. Then I read samples in both directions: 42 of 42 random warnings stated the doubt plainly, and none of 20 discussion-only reports or 11 reports with no going-concern language turned out to hide one when I read them in full.</p> <h2> 2. Blank-check companies swamp the count </h2> <p>SPACs (SIC code 6770) must complete a merger by a deadline or return the trust money, and the cash they hold outside the trust is small, so most carry the warning once the deadline is within a year: 207 of 301 in my data. That says little about shareholder risk, because the money sits in the trust. Leaving them in would add a couple of hundred warnings that mean something else entirely, so I counted them separately.</p> <h2> 3. The SEC's exchange file merges two exchanges </h2> <p>The exchange for each company came from the SEC's <code>company_tickers_exchange.json</code>. It's the best free mapping I know of, but its "NYSE" label also covers NYSE American: Cheniere Energy Partners, Uranium Energy and United States Antimony, all NYSE American listings, show up as "NYSE". For a Nasdaq vs NYSE comparison that matters, since NYSE American is home to a lot of small companies. I labeled the group "NYSE and NYSE American" rather than pretend I could split it.</p> <p>The same file is a good completeness check. 532 companies that filed a 10-Q weren't in my data, and 501 of them had no ticker in that file at all (debt-only issuers, non-traded funds). Worth checking before you call a dataset "every filer".</p> <h2> 4. Prices for OTC names can be years stale </h2> <p>For the size breakdown I multiplied shares outstanding by the latest close. For one OTC company, Creatd, the latest close in my price table was $1,850, from September 2022, which made it a $6 billion company on paper. Rather than patch names one at a time, I limited the size analysis to exchange-listed companies and used the October 26 close.</p> <p>Share classes need the same care as in any market-cap calculation: an Up-C company's total share count can be more than double its listed Class A count.</p> <h2> 5. "No later report" doesn't mean "dropped the warning" </h2> <p>To see whether warnings persist, I followed the 780 companies that warned in a 10-Q filed from January to May. Comparing each with its next 10-Q loses everyone who doesn't have one. The EDGAR quarterly form index fills the gap:</p> <ul> <li>46 filed a 10-K instead, because their fiscal year ended in between.</li> <li>15 filed only an NT 10-Q or NT 10-K, the notice that a report will be late.</li> <li>11 filed a Form 15 to stop reporting altogether.</li> </ul> <p>Of the 671 with a later 10-Q I could read, 632 still carried the warning.</p> <h2> 6. A filing is a snapshot </h2> <p>Summit Therapeutics, worth $11.6 billion, wrote the warning in its report for the quarter to June 30. Two months later, AstraZeneca agreed to buy $2.0 billion of its convertible preferred stock. The warning was true when it was written and may well be gone from the next report. If you build anything on going-concern status, store the filing date with it and check the 8-Ks filed since.</p> <h2> Where it ended up </h2> <p>The exchange comparison, size and industry breakdowns and the follow-up are in the study, with accession numbers for every company named:</p> <p><a href="https://equibles.com/research/one-in-six-nasdaq-companies-doubts-it-can-last-a-year" rel="noopener noreferrer">https://equibles.com/research/one-in-six-nasdaq-companies-doubts-it-can-last-a-year</a></p> <p>If you've worked with going-concern data: do you classify from the 10-Q text, from the auditor's opinion in the 10-K, or from a vendor flag, and how do you handle the conditional sentences?</p> <p><em>Data: SEC 10-Q filings and EDGAR indexes, processed by <a href="https://equibles.com" rel="noopener noreferrer">Equibles</a>.</em></p> datascience finance python opendata Daniel Oliveira Mon, 02 Nov 2026 11:26:51 +0000 https://dev.to/daniel_oliveira_4bd55d35d/not-every-short-is-a-bet-i-checked-the-filings-behind-the-25-most-shorted-sp-500-stocks-3h2n https://dev.to/daniel_oliveira_4bd55d35d/not-every-short-is-a-bet-i-checked-the-filings-behind-the-25-most-shorted-sp-500-stocks-3h2n <p>Kimberly-Clark is one of the most heavily shorted stocks in the S&amp;P 500. At the October 15, 2026 settlement, 50.6 million of its shares were sold short, 15.2% of its shares outstanding, up from 8.3 million a year earlier. Read the usual way, that is a growing crowd betting the Kleenex maker will fall.</p> <p>The timing says something else. The position started climbing in November 2025, right after Kimberly-Clark agreed to buy Kenvue and pay partly in its own shares. In a stock deal, arbitrage funds commonly buy the target and short the buyer, so the short position grows with the deal, not with a view on the company.</p> <p>So I checked how many of the most-shorted S&amp;P 500 stocks have a reason like that written into their own filings.</p> <h2> How I checked </h2> <ul> <li>I took FINRA's short interest for every S&amp;P 500 member at the October 15, 2026 settlement and ranked the 25 highest as a share of shares outstanding.</li> <li>For each, I searched SEC filings since October 2025 for the four standard sources of hedged short selling: a stock-for-stock deal or exchange offer, convertible or exchangeable notes, forward sales of new shares, and mandatory convertible preferred stock.</li> <li>I read every hit. Boilerplate mentions, such as debt covenants that refer to exchange offers in general, don't count.</li> </ul> <h2> The top ten </h2> <div class="table-wrapper-paragraph"><table> <thead> <tr> <th>Rank</th> <th>Stock</th> <th>Short interest</th> <th>Change in a year</th> <th>Days to cover</th> <th>Hedging source in filings</th> </tr> </thead> <tbody> <tr> <td>1</td> <td>Norwegian Cruise Line</td> <td>19.4%</td> <td>+168%</td> <td>4.2</td> <td>Exchangeable notes</td> </tr> <tr> <td>2</td> <td>Lululemon</td> <td>15.4%</td> <td>+103%</td> <td>3.7</td> <td>None found</td> </tr> <tr> <td>3</td> <td>Kimberly-Clark</td> <td>15.2%</td> <td>+508%</td> <td>10.2</td> <td>Pending stock deal (Kenvue)</td> </tr> <tr> <td>4</td> <td>Charter</td> <td>14.6%</td> <td>+12%</td> <td>5.9</td> <td>None found</td> </tr> <tr> <td>5</td> <td>Akamai</td> <td>13.0%</td> <td>+85%</td> <td>3.7</td> <td>Convertible notes</td> </tr> <tr> <td>6</td> <td>Super Micro Computer</td> <td>12.2%</td> <td>-5%</td> <td>2.6</td> <td>Convertibles, mandatory convertible preferred</td> </tr> <tr> <td>7</td> <td>T. Rowe Price</td> <td>12.0%</td> <td>+132%</td> <td>12.3</td> <td>None found</td> </tr> <tr> <td>8</td> <td>Fair Isaac</td> <td>11.2%</td> <td>+154%</td> <td>1.7</td> <td>None found</td> </tr> <tr> <td>9</td> <td>Reddit</td> <td>11.2%</td> <td>+13%</td> <td>5.1</td> <td>None found</td> </tr> <tr> <td>10</td> <td>Medtronic</td> <td>10.9%</td> <td>+830%</td> <td>2.9</td> <td>Exchange offer (MiniMed)</td> </tr> </tbody> </table></div> <p>Across the full top 25, 10 have a hedging source in their filings and 15 don't.</p> <h2> What the hedges look like </h2> <p><strong>Stock deals.</strong> Skyworks shows the full cycle: its short position rose from 10.6 million shares to 34.6 million while its Qorvo deal was pending, then fell to 22.3 million after it closed.</p> <p><strong>Exchange offers.</strong> Medtronic's short position went from 12.1 million shares at the end of August to 139.0 million on October 15, as its offer to swap shares of its MiniMed diabetes business closed 5.4 times oversubscribed.</p> <p><strong>Convertible notes.</strong> Funds that buy convertibles commonly short the stock to hedge. Akamai has $7.64 billion outstanding in five series, Super Micro $4.73 billion and Live Nation $3.5 billion.</p> <p><strong>Forward share sales.</strong> Utilities raise equity this way, and a bank borrows and sells the shares up front. Alliant Energy's 10-Q says its counterparties "borrowed and sold" its shares under the program. Its short position nearly doubled in a year, and because utility shares trade little, it takes almost ten days to cover.</p> <h2> A checklist before reading short interest as a bet </h2> <ol> <li>A pending stock deal or exchange offer: an 8-K, S-4 or Schedule TO.</li> <li>Convertible or exchangeable notes: the debt note in the latest 10-Q.</li> <li>Forward share sales: the equity note in a 10-Q, common at utilities.</li> <li>Mandatory convertible preferred stock: often on the 10-Q cover.</li> <li>The share count behind the percentage. A company that just issued shares in a deal looks more shorted than it is until its next report.</li> </ol> <h2> What this can't tell you </h2> <ul> <li>FINRA's data says how many shares are short, not who holds them or why. A documented hedging source is a reason for short selling, not proof that a given share of the position is a hedge.</li> <li>"None found" covers only those four sources. Shorts can also hedge options or ETF creations, which don't show up in company filings.</li> </ul> <p>For the 15 with no hedging source, the filings give no other explanation, and it pays to look at what changed in the past year. Lululemon's short interest doubled, and Fair Isaac's rose 154%.</p> <p>The full study has all 25 stocks, the filing sources and the method: <a href="https://equibles.com/research/not-every-short-is-a-bet-the-s-p-500s-most-shorted-stocks" rel="noopener noreferrer">Not Every Short Is a Bet: The S&amp;P 500's Most-Shorted Stocks</a>.</p> <p>Data: FINRA short interest and SEC filings, processed by Equibles.</p> datascience finance investing analytics Daniel Oliveira Mon, 02 Nov 2026 11:23:00 +0000 https://dev.to/daniel_oliveira_4bd55d35d/is-ai-costing-jobs-outside-tech-i-read-4335-sp-500-earnings-calls-46of https://dev.to/daniel_oliveira_4bd55d35d/is-ai-costing-jobs-outside-tech-i-read-4335-sp-500-earnings-calls-46of <p>I read 4,335 earnings calls from the S&amp;P 500 companies outside the technology sector, held from January 2024 to 23 October 2026, looking for one thing: an executive saying AI was changing the company's headcount.</p> <p>It happened on 41 calls, by 26 companies. That is about one call in a hundred.</p> <h2> How I counted </h2> <ul> <li>I searched every sentence that mentions AI for a link to jobs, headcount or hiring, then read each candidate by hand.</li> <li>A call counts when an executive said AI is reducing the company's headcount, or will, or that it lets the company keep headcount flat or grow it more slowly than the business.</li> <li>A broader count adds calls where AI is named next to job cuts, a hiring slowdown or a flatter structure without being given as the cause.</li> <li>I also counted denials, where an executive said the company is not using AI to cut jobs, and analyst questions about AI and headcount.</li> </ul> <p>A keyword filter would not get close. Executives mentioned AI on 61.2% of these calls in July to September 2026, and on most of them AI is a product or productivity story with no word about staff.</p> <h2> Rare, but rising </h2> <div class="table-wrapper-paragraph"><table> <thead> <tr> <th>Year</th> <th>Non-tech calls</th> <th>AI tied to headcount</th> <th>Share</th> <th>Broader count</th> <th>Share</th> </tr> </thead> <tbody> <tr> <td>2024</td> <td>1,576</td> <td>8</td> <td>0.51%</td> <td>12</td> <td>0.76%</td> </tr> <tr> <td>2025</td> <td>1,526</td> <td>16</td> <td>1.05%</td> <td>31</td> <td>2.03%</td> </tr> <tr> <td>2026, to 23 October</td> <td>1,233</td> <td>17</td> <td>1.38%</td> <td>45</td> <td>3.65%</td> </tr> <tr> <td>All</td> <td>4,335</td> <td>41</td> <td>0.95%</td> <td>88</td> <td>2.03%</td> </tr> </tbody> </table></div> <p>The jump came in the second half of 2025, when the share tripled.</p> <h2> Mostly hiring less, not layoffs </h2> <p>Of the 41 calls, 22 were about growing without adding people and 5 about not replacing people who leave. Only 14 described cut roles or lower headcount.</p> <p>Banks describe the attrition version most plainly. A large bank loses thousands of people a year, so hiring fewer of them shrinks the workforce without a layoff. Bank of America's chief executive called it "let attrition be your friend" in April 2026.</p> <p>The cuts that do get named are often small and specific. United Airlines' chief financial officer said in October 2025 that "our management headcount is 4% lower than last year", with another 4% planned for 2026.</p> <h2> Banks and a freight broker </h2> <p>Financial services companies made 20 of the 41 claims, on 2.58% of their calls. Industrials made 15, and eight of those came from one company, C.H. Robinson, which tied AI to headcount on 8 of its 11 calls and reported average headcount down 12.3% from a year earlier in Q1 2026.</p> <p>Five sectors made no such claim at all: real estate, utilities, consumer defensive, basic materials and energy.</p> <h2> The denials are growing too </h2> <p>Executives said they were not using AI to cut jobs on no call in 2024, 7 in 2025 and 13 in 2026, against 17 claims this year. One consulting chief at Marsh McLennan said there are "a lot of shenanigans going on" in companies crediting AI for headcount changes.</p> <p>Analysts are asking more often: 1 call in 2024, 6 in 2025 and 18 in 2026.</p> <h2> What this can't tell you </h2> <ul> <li>It counts what executives say on calls, not payroll data. A company can cut jobs without naming AI, and name AI for cuts that have other causes.</li> <li>2026 is a partial year, through 23 October, with the October reporting season still under way.</li> </ul> <p>The full study has the sector table, the companies that named cuts and every quote with its source: <a href="https://equibles.com/research/executives-outside-tech-say-ai-means-hiring-less-not-layoffs" rel="noopener noreferrer">Executives Outside Tech: AI Means Hiring Less, Not Layoffs</a>.</p> <p>Data: S&amp;P 500 earnings call transcripts, processed by Equibles.</p> ai career datascience analytics Daniel Oliveira Mon, 02 Nov 2026 00:18:45 +0000 https://dev.to/daniel_oliveira_4bd55d35d/i-read-every-ai-claim-on-4335-earnings-calls-a-keyword-filter-would-have-got-most-of-them-wrong-2mlo https://dev.to/daniel_oliveira_4bd55d35d/i-read-every-ai-claim-on-4335-earnings-calls-a-keyword-filter-would-have-got-most-of-them-wrong-2mlo <p>Apollo's chief economist argued in late October that AI has delivered "zero financial payoff" to S&amp;P 500 companies outside tech, based on flat margins and revenue per worker. I wanted the firm-level version of that question: when executives outside tech talk about AI on their earnings calls, how often do they put a number on what it delivered?</p> <p>The sample was every earnings call held since January 2024 by S&amp;P 500 companies outside the tech sector, 4,335 calls from speaker-by-speaker transcripts. The headline:</p> <div class="table-wrapper-paragraph"><table> <thead> <tr> <th>Calls held</th> <th>Executive mentioned AI</th> <th>Put a number on AI</th> </tr> </thead> <tbody> <tr> <td>Jan to Mar 2024</td> <td>33.8%</td> <td>2.1%</td> </tr> <tr> <td>Jul to Sep 2025</td> <td>54.7%</td> <td>5.8%</td> </tr> <tr> <td>Jul to Sep 2026</td> <td>61.2%</td> <td>9.0%</td> </tr> </tbody> </table></div> <p>Counting mentions is easy. Counting "put a number on AI" is not, and the gap between what a filter finds and what a person reading the sentence finds was the most useful lesson of the project.</p> <h2> 1. The obvious filter has about 35% precision </h2> <p>The first pass looked for sentences by company speakers (not analysts) that mention AI and contain a number and a benefit word in the same or the next sentence:<br> </p> <div class="highlight js-code-highlight"> <pre class="highlight python"><code><span class="kn">import</span> <span class="n">re</span> <span class="n">AI</span> <span class="o">=</span> <span class="n">re</span><span class="p">.</span><span class="nf">compile</span><span class="p">(</span> <span class="sa">r</span><span class="sh">"</span><span class="s">(?&lt;![A-Za-z])A\.?I\.?(?![A-Za-z])</span><span class="sh">"</span> <span class="c1"># "AI" or "A.I.", not "AIG" or "said" </span> <span class="sa">r</span><span class="sh">"</span><span class="s">|(?i:\bartificial intelligence\b|\bagentic\b|\bgen ?AI\b|\bLLMs?\b|\bmachine learning\b)</span><span class="sh">"</span> <span class="p">)</span> <span class="n">NUM</span> <span class="o">=</span> <span class="n">re</span><span class="p">.</span><span class="nf">compile</span><span class="p">(</span><span class="sa">r</span><span class="sh">"</span><span class="s">\$\s?\d[\d,.]*|\b\d[\d,.]*\s?(%|(percent|bps|million|billion|hours?|days?|times|x)\b)</span><span class="sh">"</span> <span class="sa">r</span><span class="sh">"</span><span class="s">|\b(half|halved|doubled|two-thirds)\b</span><span class="sh">"</span><span class="p">,</span> <span class="n">re</span><span class="p">.</span><span class="n">I</span><span class="p">)</span> <span class="n">BENEFIT</span> <span class="o">=</span> <span class="n">re</span><span class="p">.</span><span class="nf">compile</span><span class="p">(</span><span class="sa">r</span><span class="sh">"</span><span class="s">\b(productiv\w*|efficienc\w*|sav\w*|reduc\w*|cut|faster|automat\w*|headcount)\b</span><span class="sh">"</span><span class="p">,</span> <span class="n">re</span><span class="p">.</span><span class="n">I</span><span class="p">)</span> <span class="k">def</span> <span class="nf">candidates</span><span class="p">(</span><span class="n">sentences</span><span class="p">):</span> <span class="k">for</span> <span class="n">i</span><span class="p">,</span> <span class="n">s</span> <span class="ow">in</span> <span class="nf">enumerate</span><span class="p">(</span><span class="n">sentences</span><span class="p">):</span> <span class="k">if</span> <span class="n">AI</span><span class="p">.</span><span class="nf">search</span><span class="p">(</span><span class="n">s</span><span class="p">):</span> <span class="n">window</span> <span class="o">=</span> <span class="n">s</span> <span class="o">+</span> <span class="sh">"</span><span class="s"> </span><span class="sh">"</span> <span class="o">+</span> <span class="p">(</span><span class="n">sentences</span><span class="p">[</span><span class="n">i</span> <span class="o">+</span> <span class="mi">1</span><span class="p">]</span> <span class="k">if</span> <span class="n">i</span> <span class="o">+</span> <span class="mi">1</span> <span class="o">&lt;</span> <span class="nf">len</span><span class="p">(</span><span class="n">sentences</span><span class="p">)</span> <span class="k">else</span> <span class="sh">""</span><span class="p">)</span> <span class="k">if</span> <span class="n">NUM</span><span class="p">.</span><span class="nf">search</span><span class="p">(</span><span class="n">window</span><span class="p">)</span> <span class="ow">and</span> <span class="n">BENEFIT</span><span class="p">.</span><span class="nf">search</span><span class="p">(</span><span class="n">window</span><span class="p">):</span> <span class="k">yield</span> <span class="n">window</span> </code></pre> </div> <p>The <code>AI</code> part is case-sensitive and allows "A.I.", the way some transcripts write it; the letter lookarounds keep it from matching inside names such as AIG.</p> <p>It produced 534 candidate passages on 343 calls. Reading them, 122 calls had a real figure for what AI delivered. The rest were things a regex can't tell apart:</p> <ul> <li> <strong>Usage, not payoff.</strong> The share of staff who finished the training, the number of prompts in a month, the share of trucks with AI tools. A number next to AI, but it measures adoption.</li> <li> <strong>Spending.</strong> A company saying how many billions it put into AI and other technology. The money goes the wrong way.</li> <li> <strong>AI as demand.</strong> Utilities, data center landlords and materials companies talk about AI as customers: "data center and AI share of U.S. electricity consumption will triple". Real numbers, not a payoff from using AI.</li> <li> <strong>Bundled programs.</strong> A sentence reporting a large cost-savings total, then saying AI will speed up efficiency from here. The number and the AI sit together but are not connected.</li> <li> <strong>Speech recognition.</strong> A medical device company's "significant AI reduction" turned out to be AHI, the apnea index, misheard.</li> </ul> <h2> 2. Then it misses about a quarter of the real ones </h2> <p>A second pass dropped the benefit-word requirement and widened it to revenue words (conversion, approval rates, growth, uplift), and listed every AI sentence with a number on calls the first pass hadn't flagged, or had flagged for a different sentence. It found 36 more calls.</p> <p>The misses were mostly revenue and customer outcomes, phrased in ways a cost-savings vocabulary doesn't catch: an AI shopping assistant whose users convert at "triple" the rate (no digit at all), approval rates up 50 basis points from AI decisioning, an AI product line adding 300 basis points of organic growth. A few were cost claims split across sentences further apart than the window.</p> <p>The lesson I'd keep: for a claim-level count, treat the regex as a way to rank what to read, not as the classifier. I read every candidate and every second-pass sentence; the labels took longer than all the code.</p> <h2> 3. Check who you filtered out as "not the company" </h2> <p>A day after publishing I found two more holes, both in the filters rather than the reading, and both in the first pass and the second.</p> <p>The first was the speaker filter. Transcripts give each turn a role, and I kept CEO, CFO, chair, COO and the like. Executives whose titles don't fit the role list, such as a chief commercial officer or a division president, are stored as "Other", and I had grouped "Other" with the operator. C.H. Robinson's results were mostly presented by those executives:</p> <blockquote> <p>"In the AI agent's first few months, it has been determining the freight class and code for about 2,000 orders a day, and it has reduced the processing time from 10 minutes or more per shipment to 10 seconds or less."</p> </blockquote> <p>The second was the window. A figure often comes first and the AI credit second. A sentence with both in it works, but "total non-standard surcharges are generating a significant year-over-year benefit of over $180 million annualized. This is the result of a new AI image capture process" puts the number one sentence before the AI. Fixing both added 385 passages to the first pass and 459 to the second, all read by hand. They added 65 calls with a figure, 223 in all, and raised last quarter's share from 6.5% to 9.0%. Turns with no role at all I only counted when reading them showed a company speaker, because some of them are analysts.</p> <h2> 4. Decide what "credit" means before you label </h2> <p>Many figures credit AI along with other things: "efficiencies from organizational realignment, process redesign through AI and automation, procurement and vendor rationalization ... more than $1.5 billion". I labeled those separately from figures credited to AI alone. Last quarter 27 of the 35 calls with a number credited AI alone; most of the shared-credit ones are company-wide cost programs. Without that split, the count mixes an AI agent that resolves a stated share of support issues with a reorganization that mentions AI once.</p> <h2> 5. Check for the same transcript on two calls </h2> <p>Before computing shares by quarter I hashed every transcript and grouped by company. Across the whole transcript archive, 182 transcripts were stored on more than one call event of the same company, 67 of them in different quarters; one company's investor webinar sat on five consecutive quarters' calls. In the S&amp;P 500 sample there were 18 such pairs. Keeping one call per company per quarter handled the same-quarter ones, and I dropped the three pairs that crossed quarters:<br> </p> <div class="highlight js-code-highlight"> <pre class="highlight sql"><code><span class="c1">-- PostgreSQL: identical transcripts on more than one call of the same company</span> <span class="k">with</span> <span class="n">h</span> <span class="k">as</span> <span class="p">(</span> <span class="k">select</span> <span class="n">call_id</span><span class="p">,</span> <span class="n">company_id</span><span class="p">,</span> <span class="n">md5</span><span class="p">(</span><span class="n">string_agg</span><span class="p">(</span><span class="nb">text</span><span class="p">,</span> <span class="s1">' '</span> <span class="k">order</span> <span class="k">by</span> <span class="n">turn_index</span><span class="p">))</span> <span class="k">as</span> <span class="n">hash</span> <span class="k">from</span> <span class="n">transcript_turns</span> <span class="k">group</span> <span class="k">by</span> <span class="n">call_id</span><span class="p">,</span> <span class="n">company_id</span> <span class="p">)</span> <span class="k">select</span> <span class="n">company_id</span><span class="p">,</span> <span class="n">hash</span><span class="p">,</span> <span class="k">count</span><span class="p">(</span><span class="o">*</span><span class="p">)</span> <span class="k">as</span> <span class="n">calls</span> <span class="k">from</span> <span class="n">h</span> <span class="k">group</span> <span class="k">by</span> <span class="n">company_id</span><span class="p">,</span> <span class="n">hash</span> <span class="k">having</span> <span class="k">count</span><span class="p">(</span><span class="o">*</span><span class="p">)</span> <span class="o">&gt;</span> <span class="mi">1</span><span class="p">;</span> </code></pre> </div> <h2> What the clean count shows </h2> <ul> <li>Executives outside tech talked about AI on 61.2% of last quarter's calls and put a number on its payoff on 9.0%, about one in seven of the calls that mentioned it.</li> <li>Eleven companies account for 68 of the 223 calls with a number since 2024.</li> <li>The most common figure is a savings target for a company-wide program; the concrete results are narrow, in customer service, claims and document processing, and code.</li> <li>Analysts asked about AI savings, productivity or returns on 6.7% of calls last quarter, against about 1% in 2024.</li> </ul> <p>The full write-up, with the charts, the sector table and every quote linked to its call: <a href="https://equibles.com/research/ai-talk-has-nearly-doubled-outside-tech-proof-of-payoff-is-still-rare" rel="noopener noreferrer">AI Talk Has Nearly Doubled Outside Tech</a></p> <p>If you've labeled claims in transcripts or other long documents, did you find a way to get the filter's precision up, or did you also end up reading everything?</p> <p>Data: S&amp;P 500 earnings call transcripts, processed by Equibles.</p> datascience python ai nlp Daniel Oliveira Sun, 01 Nov 2026 04:17:28 +0000 https://dev.to/daniel_oliveira_4bd55d35d/does-congrats-on-the-quarter-mean-anything-i-read-the-analyst-questions-on-59306-earnings-calls-bm3 https://dev.to/daniel_oliveira_4bd55d35d/does-congrats-on-the-quarter-mean-anything-i-read-the-analyst-questions-on-59306-earnings-calls-bm3 <p>Analysts open their earnings call questions with "congrats on the quarter" so often that it sounds like a reflex. I wanted to know whether it carries any information, so I read the analyst questions on 59,306 earnings calls held by 3,562 US-listed companies from 2020 to 2025, sorted every congratulation by what it was for, and lined the calls up against the stock's move against SPY.</p> <h2> The short answer </h2> <div class="table-wrapper-paragraph"><table> <thead> <tr> <th>Analysts who praised the results</th> <th>Calls</th> <th>Median return vs SPY, to the next day's close</th> <th>Beat SPY</th> </tr> </thead> <tbody> <tr> <td>None (no congratulations at all)</td> <td>35,833</td> <td>-0.7%</td> <td>44.3%</td> </tr> <tr> <td>One</td> <td>11,199</td> <td>+1.0%</td> <td>56.7%</td> </tr> <tr> <td>Two</td> <td>4,206</td> <td>+1.7%</td> <td>60.4%</td> </tr> <tr> <td>Three or more</td> <td>2,961</td> <td>+2.6%</td> <td>64.6%</td> </tr> </tbody> </table></div> <p>The pleasantry is informative, up to a point. When several analysts praise the results, the stock has usually moved the same way. It is not a forecast.</p> <h2> Five findings </h2> <p><strong>1. On most calls, nobody says it.</strong> On 60.4% of calls not one analyst congratulated anybody. Counting each analyst once per call, 12.7% of the 351,985 analysts congratulated someone at the start of a question. "Congrats" (24,797 analysts) beat "congratulations" (20,083).</p> <p><strong>2. Not every congratulation is about the quarter.</strong> Many are about a promotion, a retirement or a deal, and they cluster. On Etsy's call on October 29, 2025, six of seven analysts opened with congratulations, only one of them about the business (the rest were about the leadership transition or for "everyone"), and the stock fell 16.1% against SPY by the next day's close. When the only congratulations were for a new role or retirement, the stock beat SPY 47.0% of the time, about the same as no congratulations at all.</p> <p><strong>3. It holds across sizes and years.</strong> Split into small, mid and large companies by trading volume, calls where two or more analysts praised the results beat SPY 61.2%, 61.4% and 63.4% of the time, against 43.7%, 44.8% and 44.3% with none. It held in every year from 2020 to 2025.</p> <p><strong>4. It's a verdict, not foresight.</strong> By the time the Q&amp;A starts, analysts have read the same release the market is trading on. Over the next 60 sessions the gap mostly closed:</p> <div class="table-wrapper-paragraph"><table> <thead> <tr> <th>Analysts who praised the results</th> <th>Median return vs SPY, next 60 sessions</th> <th>Beat SPY</th> </tr> </thead> <tbody> <tr> <td>None</td> <td>-2.6%</td> <td>43.2%</td> </tr> <tr> <td>One</td> <td>-2.3%</td> <td>44.5%</td> </tr> <tr> <td>Two</td> <td>-1.2%</td> <td>47.1%</td> </tr> <tr> <td>Three or more</td> <td>-1.4%</td> <td>47.5%</td> </tr> </tbody> </table></div> <p>That small edge is in line with post-earnings drift, and with Milian and Smith's earlier study of analyst praise on calls from 2003 to 2013. A 47.5% hit rate is not something to trade on.</p> <p><strong>5. "Great question" carries nothing.</strong> An executive opened an answer with "great question" or "good question" on 38.9% of calls. Those stocks beat SPY by the next day's close 49.5% of the time; on calls without it, 49.1%.</p> <h2> The traps I hit </h2> <ul> <li> <strong>Classify the congratulation, not the call.</strong> I took the sentence containing "congrat" in the first 400 characters of an analyst turn and sorted it into people, results, deals or not specified. A sentence that mentions both a person and the results counts as people. A random sample of 25 sorted as results were all about the quarter, results or progress; 20 of 25 sorted as people were about a person.</li> <li> <strong>Use a window that covers both release times.</strong> Returns run from the close before the call to the close of the day after it, minus SPY, so the window covers results released before the open or after the close.</li> <li> <strong>Praise and price share a cause.</strong> Analysts and the market react to the same numbers, and by the time analysts speak they have usually seen the first part of the move. Neither drives the other.</li> <li> <strong>Who gets called on matters.</strong> Companies choose which analysts ask questions, and Cohen, Lou and Malloy found some favor friendly ones.</li> <li> <strong>Survivors only.</strong> Companies no longer listed are mostly missing from the transcript set.</li> </ul> <p>The full study, with the charts by year and by industry, the method and sources, is on <a href="https://equibles.com/research/what-congrats-on-the-quarter-tells-you-about-a-stock" rel="noopener noreferrer">Equibles</a>, where the transcripts come from.</p> <p>Data: earnings call transcripts and daily prices, processed by Equibles. This is not investment advice.</p> datascience finance investing nlp Daniel Oliveira Sun, 01 Nov 2026 04:13:57 +0000 https://dev.to/daniel_oliveira_4bd55d35d/what-happens-after-a-nasdaq-1-warning-i-followed-145-companies-for-up-to-22-months-i0f https://dev.to/daniel_oliveira_4bd55d35d/what-happens-after-a-nasdaq-1-warning-i-followed-145-companies-for-up-to-22-months-i0f <p>When a Nasdaq stock closes with a bid under $1.00 for 30 business days in a row, the exchange sends a deficiency letter and the company has 180 days to get back above $1 for ten straight closes. I wanted to know what that letter actually leads to, so I followed every Nasdaq company that disclosed its first bid-price notice in the first half of 2025, 145 of them, to October 26, 2026.</p> <h2> The short answer </h2> <div class="table-wrapper-paragraph"><table> <thead> <tr> <th>Outcome by October 26, 2026</th> <th>Companies</th> </tr> </thead> <tbody> <tr> <td>Still listed on a US exchange</td> <td>108 (74.5%)</td> </tr> <tr> <td>of which trading at $1 or more</td> <td>78</td> </tr> <tr> <td>of which back under $1</td> <td>30</td> </tr> <tr> <td>Delisted, suspended or deregistered</td> <td>19</td> </tr> <tr> <td>Acquired</td> <td>11</td> </tr> <tr> <td>Bankrupt</td> <td>7</td> </tr> </tbody> </table></div> <p>Most companies stay listed. Most of them get there with a reverse split. And most people who held the stock on the day of the letter are worse off.</p> <h2> Five findings </h2> <p><strong>1. A reverse split is the usual way out, and it comes late.</strong> 85 of the 145 (58.6%) did at least one reverse split after the letter, typically 1-for-20. The median company split 258 days after disclosing the letter, well into the second 180-day period. 14 did more than one.</p> <p><strong>2. Nearly half recovered on their own at least once.</strong> 59 of the 126 companies with price history (46.8%) closed at $1 or more for ten straight days before any split, 46 of them inside the first 180 days. But 24 of those 59 later got a new letter.</p> <p><strong>3. The price on the day of the letter mattered most.</strong></p> <div class="table-wrapper-paragraph"><table> <thead> <tr> <th>Price when the letter was disclosed</th> <th>Companies</th> <th>Back above $1 before any split</th> <th>Delisted or bankrupt</th> </tr> </thead> <tbody> <tr> <td>Under $0.50</td> <td>23</td> <td>4</td> <td>8</td> </tr> <tr> <td>$0.50 to $0.79</td> <td>73</td> <td>37</td> <td>8</td> </tr> <tr> <td>$0.80 or more</td> <td>30</td> <td>18</td> <td>2</td> </tr> </tbody> </table></div> <p><strong>4. Staying listed is not the same as getting whole.</strong> Among the 107 still-listed companies with prices on both dates, the median stock was down 47.3% from the close on the day the letter was disclosed, against a 32.4% gain for SPY over the same windows. The 18 delisted or bankrupt companies with price data lost a median 97.7% by their last close.</p> <p><strong>5. Two notices at once was the worst combination.</strong> 58 companies also received a notice for another listing rule, most often stockholders' equity or a late filing. 27.6% of those were delisted or went bankrupt, against 11.5% of the companies that only had the price problem.</p> <h2> The traps I hit </h2> <ul> <li> <strong>Not every Item 3.01 8-K is a first letter.</strong> I read every notice and removed 11 that were extensions, hearing decisions, a bankruptcy notice, or companies that got no compliance period at all because of earlier reverse splits.</li> <li> <strong>Split-adjust everything.</strong> Vivakor did a 1-for-200, a 1-for-20 and a 1-for-15 split in 2026, so 60,000 old shares became one. Raw prices make it look like a recovery.</li> <li> <strong>Survivorship cuts the other way here.</strong> 19 companies have no price history in my data, 18 of them already acquired, bankrupt or delisted, so the recovery rates lean toward the survivors.</li> <li> <strong>A split is a marker, not a cause.</strong> Companies that split were down a median 83.0% and those that didn't were up 41.2%, but that is because the ones that recovered never needed a split.</li> </ul> <h2> How to check a company yourself </h2> <ol> <li>Find the 8-K under Item 3.01: it gives the letter date and the deadline.</li> <li>A second 180-day grant means the company has told Nasdaq it will do a reverse split if needed.</li> <li>Look for a proxy asking shareholders to approve a split, usually with a wide range of ratios.</li> <li>Check the last split: since January 2025, a company that falls below $1 within a year of a reverse split gets no compliance period at all.</li> <li>Look for a second deficiency, such as stockholders' equity, on top of the price notice.</li> </ol> <p>The full study, with the method, every source filing and the charts, is on <a href="https://equibles.com/research/nasdaqs-1-warning-most-companies-stay-listed-most-shares-fall" rel="noopener noreferrer">Equibles</a>, where I read the 8-Ks and price history.</p> <p>Data: SEC filings and daily prices, processed by Equibles. This is not investment advice.</p> datascience finance investing stocks Daniel Oliveira Sun, 01 Nov 2026 02:36:20 +0000 https://dev.to/daniel_oliveira_4bd55d35d/alphabet-and-meta-bought-back-0-of-stock-in-2026-i-read-five-hyperscaler-10-qs-to-see-where-the-3nhe https://dev.to/daniel_oliveira_4bd55d35d/alphabet-and-meta-bought-back-0-of-stock-in-2026-i-read-five-hyperscaler-10-qs-to-see-where-the-3nhe <p>Last week's selloff in Alphabet, Meta and Amazon was mostly told as a free cash flow story. I wanted to see the other side of the same cash flow statement: what happened to buybacks. So I read the latest 10-Q from each of the five big AI spenders and pulled the repurchase lines.</p> <p>The short version: Alphabet and Meta, which bought back $62.2B and $30.1B of their own stock in 2024, repurchased nothing at all in the first nine months of 2026.</p> <h2> The table </h2> <div class="table-wrapper-paragraph"><table> <thead> <tr> <th>Company</th> <th>Period</th> <th>Buybacks this year</th> <th>Same period a year earlier</th> <th>Authorization left</th> <th>Capex this year</th> <th>Capex a year earlier</th> </tr> </thead> <tbody> <tr> <td>Alphabet</td> <td>Jan to Sep 2026</td> <td>$0</td> <td>$39.8B</td> <td>$69.5B of $70.0B</td> <td>$132.0B</td> <td>$63.6B</td> </tr> <tr> <td>Meta</td> <td>Jan to Sep 2026</td> <td>$0</td> <td>$26.2B</td> <td>$25.0B</td> <td>$83.1B</td> <td>$48.3B</td> </tr> <tr> <td>Microsoft</td> <td>Jul to Sep 2026</td> <td>$2.9B</td> <td>$4.0B</td> <td>$37.7B of $60.0B</td> <td>$39.4B</td> <td>$19.4B</td> </tr> <tr> <td>Amazon</td> <td>Jan to Sep 2026</td> <td>$0</td> <td>$0</td> <td>$6.1B of $10.0B</td> <td>$175.7B</td> <td>$92.3B</td> </tr> <tr> <td>Oracle</td> <td>Jun to Aug 2026</td> <td>$0</td> <td>$93M</td> <td>$6.3B</td> <td>n/a</td> <td>n/a</td> </tr> </tbody> </table></div> <p>Capex here is the cash flow statement line ("purchases of property and equipment", "additions to property and equipment" at Microsoft). Microsoft's fiscal year starts in July and Oracle's in June, so their periods are single quarters. Microsoft's buyback figure is the repurchase program only; it excludes shares withheld for employee taxes.</p> <h2> What stands out </h2> <p><strong>Alphabet switched the buyback off, it did not trim it.</strong> The 10-Q says it plainly: "In the three and nine months ended September 30, 2026, there were no repurchases of the company's Class A or Class C shares." The $70.0B authorization from April 2025 still has $69.5B left. Over the same nine months, capex roughly doubled, from $63.6B to $132.0B.</p> <p><strong>Meta is borrowing instead of buying.</strong> Meta "did not repurchase any shares of Class A common stock during the nine months ended September 30, 2026", against $26.2B a year earlier. Its largest financing inflow was $24.9B of net proceeds from notes sold in May 2026. Capex rose from $48.3B to $83.1B.</p> <p><strong>Microsoft is the only one still buying, and at a slower pace.</strong> It repurchased 6 million shares for $2.9B in the September quarter, down from 8 million shares for $4.0B a year earlier, while quarterly capex doubled to $39.4B.</p> <p><strong>Amazon never relied on buybacks</strong>, so its story is the capex line alone: $175.7B in nine months, against $92.3B a year earlier. On Amazon's own definition, trailing twelve month free cash flow went from $14.8B to negative $42.4B.</p> <p><strong>Oracle's program is idle too</strong>: no repurchases in the quarter ended August 31, 2026, against $93M a year earlier.</p> <p>Put together, Alphabet and Meta went from $66.1B of buybacks in the first nine months of 2025 to zero in the same months of 2026, while Alphabet, Meta and Amazon together spent $390.7B on capex, 1.9 times the $204.2B of a year earlier.</p> <h2> Why it matters for anyone screening stocks </h2> <p>Buybacks were a steady bid under Alphabet and Meta for years. A company that stops buying is not a sign of trouble on its own, and both authorizations are still open, so either can restart at any time. But it does change who is buying the stock on a bad day, which is worth keeping in mind when reading this season's post-earnings moves.</p> <p>It also shows why I like reading the filing instead of a headline number. "Negative free cash flow" sounds the same for all five companies, but the 10-Qs show five different funding choices: stop buybacks, issue notes, keep buying, never buy, or pause.</p> <h2> How I pulled it </h2> <p>Every number above comes from the companies' own 10-Q filings. I searched each filing for "repurchase" and for the capex line, then read the cash flow statement and the stockholders' equity note. If you want to check a line yourself, these are the filings:</p> <ul> <li><a href="https://equibles.com/stocks/googl/documents/c3896b76-d98f-4bc7-80fb-2c0fb2426697" rel="noopener noreferrer">Alphabet 10-Q, quarter ended September 30, 2026</a></li> <li><a href="https://equibles.com/stocks/meta/documents/16c3976d-4e32-4d23-89ef-b9e0a4ae3c04" rel="noopener noreferrer">Meta 10-Q, quarter ended September 30, 2026</a></li> <li><a href="https://equibles.com/stocks/msft/documents/007b385b-930e-472d-aa05-d70a9cfa50ed" rel="noopener noreferrer">Microsoft 10-Q, quarter ended September 30, 2026</a></li> <li><a href="https://equibles.com/stocks/amzn/documents/5e49ac8c-c72f-4bb7-9729-23c45d8fafd1" rel="noopener noreferrer">Amazon 10-Q, quarter ended September 30, 2026</a></li> <li><a href="https://equibles.com/stocks/orcl/documents/03a64054-321b-47fb-9e1b-7127badfcc73" rel="noopener noreferrer">Oracle 10-Q, quarter ended August 31, 2026</a></li> </ul> <p>I did the searching through <a href="https://equibles.com" rel="noopener noreferrer">Equibles</a>, which lets you search inside a single SEC filing and jump to the exact line, from the site or from ChatGPT and Claude.</p> <p>Data: SEC filings, processed by Equibles.</p> investing stocks datascience finance