This article was written by Miles Grauberd – Investment Analyst at I Know First.

Where Performance Actually Comes From

A large cap manager benchmarked to the S&P 500 may hold eighty positions, but only a handful of decisions determine the year. How much technology to hold relative to the index. Which sector funds that position. How much exposure to carry in the largest companies at the top of the benchmark.

Index concentration is why. The S&P 500 closed the first half of 2026 with 503 constituents and a market capitalization above $68 trillion, and information technology accounts for close to a third of that. When one sector and seven companies represent that share of the index, a risk model run on a typical active book will attribute more tracking error to sector and megacap positioning than to the several hundred individual stock bets beneath them.

This sets the standard a forecast has to meet. Ranking a thousand small caps addresses a part of the portfolio with little influence on the outcome. Ranking the eleven GICS Level 1 sectors and the largest index constituents addresses the part that determines it.

The Mechanics of Active Weight

Institutional equity portfolios run against a benchmark. Every holding carries two weights, one in the index and one in the portfolio, and the difference between them is the active weight. That difference is the only figure that affects performance measured against the index.

A name at 7% of the S&P 500 held at 9% carries a 200 basis point overweight. If the stock beats the index by 10% over the holding period, the position contributes roughly 20 basis points of relative return. The same name held at 4% carries a 300 basis point underweight, a substantial short position in relative terms even though the manager owns the stock.

Active weights sum to zero. Every overweight is funded by an underweight elsewhere, so a view on technology is also a view on whatever gets sold to pay for it. Portfolios assembled by adding conviction names without deciding the funding source accumulate bets no one selected.

Long only mandates face an asymmetry. The maximum overweight is set by risk limits. The maximum underweight is the benchmark weight itself, since zero is the floor. A megacap at 7% of the index permits a 700 basis point underweight. A constituent at 4 basis points permits nothing meaningful. Negative views on small index members have no route into a long only portfolio, which concentrates institutional research at the top of the index.

Three measures govern how far a manager can go. Active share measures how much of the portfolio differs from the index. Tracking error, the annualized volatility of active return, sets the risk budget, and typical mandates run between 2% and 6%. Information ratio, active return divided by tracking error, measures the result. Position sizes therefore follow from the risk budget rather than from conviction, because a manager working inside a 3% tracking error limit cannot spend it on a single idea.

What The Algorithm Publishes

I Know First runs a predictive algorithm built on artificial intelligence and machine learning, applied each trading day across roughly 1,500 tickers, including all eleven GICS Level 1 sector ETFs and the largest constituents of the S&P 500. The methodology is described on the I Know First website.

Every ticker carries two numbers across six horizons. The signal reports strength and direction. The predictability score, bounded between zero and one, reports how reliably the model has read that ticker in the past. The output arrives before the open, every day, with no revision cycle and no narrative attached. Sector views inside most institutions update on a committee calendar. The algorithm updates on the market’s.

What a Disagreement Looks Like

A forecast that agrees with the price adds nothing a chart does not already show. The information sits in the disagreements, and they take a specific shape. The signal moves while the price keeps doing what it was already doing, nothing confirms the reading for days or weeks, and then the price turns in the direction the signal had been pointing. During the gap the algorithm looks wrong, because early and wrong are indistinguishable until the price settles the question.

That gap is what an active weight framework is built to hold. A manager acting on a binary buy or sell instruction gets punished for being early. A manager adjusting a weight does not. For a single large position, three states govern the weight.

Signal turns up sharply from a low base while the price is flat: move to overweight. Signal and price both advancing: hold. Signal fading while the price keeps making highs: cut back to benchmark weight, and go underweight if the gap widens.

Nothing there asks a manager to sell a position or to call a top. Each state is a weight, every move between them is incremental, and the size of each step comes from the risk budget rather than from conviction. Every figure below comes from the daily forecast archive between March and July 2026, recorded as the files arrived, with no revision.

Where The Weight Sits

Institutions spent the first half of 2026 making exactly these decisions, and making them in size. Goldman Sachs data put the Magnificent Seven at roughly 14.5% of total US hedge fund exposure by June, close to a three year low, after the steepest six month reduction since the 2022 bear market. The active weight call was the trade of the half year. The only question was who made it early.

Technology: the signal stopped confirming the rally

From the March low to the end of May, XLK gained 49.8% and the signal went with it, climbing from 82.0 to 341.9, a rise of 317%. For two months the two moved together and there was nothing to act on.

What changed first was the character of the signal, not its level. It began reacting disproportionately to minor setbacks, falling sharply on pullbacks that barely registered on the chart before recovering just as quickly. Through May it oscillated between 155.0 and 327.1, reversing direction almost weekly and never sustaining a trend. A price advancing in orderly fashion while its signal swings through a two to one range is a conviction problem, and it was apparent in the daily file well before it surfaced in the price.

The cracks opened in early June. Broadcom beat on revenue and earnings on June 3 but guided third quarter AI chip sales below expectations and declined to raise its full year forecast, and the stock fell 14% the next day. The algorithm did not wait. The signal dropped from 341.9 on June 3 to 155.3 on June 7, losing 55% in four sessions, and it never returned to those highs, peaking at 250.3 through all of July. XLK had topped at 197.97 on June 2 and spent the following weeks fighting for gains it could not hold before falling 11.30%. Every Magnificent Seven name finished June in the red as the market stopped paying for AI capital spending.

Financials: strength with nothing to show for it

Through late March and April the XLF signal was persistently strong and rising, punctuated by sharp spikes, printing 118.6 on March 24, 130.7 on April 17 and 189.1 on April 22, while the sector went from 48.66 to 51.95 and nobody wanted to talk about banks. Sustained strength against a price that will not move is the buy configuration. XLF reached 56.75 by July 16, up 13.48% from the March low, printed a golden cross on July 9 for the first time since December 2023, and gained more than 8% from the start of June while the S&P 500 went sideways. Every one of those confirmations arrived after the signal had been strong for two months.

By the end of July the position warranted review. The price had stalled since the July 16 high, and the signal came off with it, dropping from 142.8 on July 23 to 96.0 the following session and 117.4 on July 26, having never regained the 189.1 it reached in April at a price 7% lower. The advance was intact and the reading still positive, so this is not the configuration that precedes a decline. It is the point at which an overweight earns less than it did, and the case for carrying the full position rests on confirmation the signal is no longer providing.

Healthcare: the signal arrived first

XLV went from 144.10 on March 20 to 148.74 on June 20, roughly 3% in three months, while the signal held elevated readings throughout. In the final week of June it accelerated from 83.8 to 166.7, and the sector ran to 164.44 by July 7. The catalyst followed: a US coverage expansion took effect on July 1, with Eli Lilly estimating some 20 million Medicare patients might newly qualify for obesity drug access. Healthcare also carried a predictability score that never rose above 0.36, and the signal was right anyway, which made a clear case for an overweight and paid well.

The case for holding it has weakened. The signal peaked at 166.7 on June 28 and has settled into a 95 to 120 band since, roughly 35% below that high, while the price stopped advancing after July 7 and has traded sideways near 161. The sector is not breaking down and the reading remains positive, but the algorithm is no longer pushing the position forward. That is the point to take profit and return the overweight to benchmark, leaving room to add again if the signal turns back up.

Communication services: still unresolved

Through late March the XLC signal sat between 9.5 and 26.3, near the floor of its range, while the price climbed to its April 17 high of 118.79. A very low signal against a rising price is the mirror image of the financials setup, and it resolved the same way. The sector broke down at the end of May and fell about 4% over the second quarter while the S&P 500 rose 14%. Earnings did the rest. Netflix guided third quarter revenue below expectations on July 16 and fell 9% over the following two sessions to a 52-week low, and Alphabet’s raised capital spending forecast drew a similar reaction days later.

The positions have now reversed. The signal reached 178.5 on July 16 and read 119.1 on July 26, while the price closed at 105.38, the low of the period. In March the signal was on the floor with the price near its high, and the sector broke. By late July the signal held in the upper half of its range with the price at its bottom, which is the configuration that preceded the advance in financials. This is no longer a benchmark weight position. It sits above neutral and moves to a full overweight if the signal continues to rise from there, with the price already discounting the earnings that caused the damage.

The Magnificent Seven

No sector weight expresses a view on these companies, because the seven span three sectors and stopped moving as one. Measured across the first half of 2026 the group fell roughly 7% while XLK gained 27.5%, a gap no single sector weight can express. Apple and Microsoft sit in the same GICS sector and went opposite directions across the four months covered here. Positions have to be taken name by name.

Apple: overweight, then neutral, then under

The signal bottomed at 68.9 on March 29, one day before Apple’s low of 246.40, then spiked to 531.7 on April 12 with the stock at 260.24. A reading seven times its March low against a price that had barely left the bottom is the entry, and it takes a manager from benchmark weight to overweight. Apple rose 29.46% from there and closed at a record 336.91 on July 27, three days before earnings, having gained about 15% in July alone and briefly carrying market value past $5 trillion and ahead of Nvidia. 

The following was written before Apple reported fiscal third quarter results on July 30. We see Apple moving to all time highs while the signal is falling, so it is a clear sign to move to equal weight from overweight, and if we see further price movement downwards we would move to underweight. The signal had eased from 271.5 on July 17 to 199.4 by July 26, roughly 62% below the April spike.

That is what happened. Apple beat on both lines, with revenue of $109.42 billion and iPhone sales up 22%, but Services missed, Greater China came up short, and management guided fourth quarter revenue growth to 9% to 11% against the 12% the street expected, citing memory shortages. The stock fell 7.35% on July 31, its worst session in 16 months, opening down 8.6% and erasing roughly $430 billion of market value. This is the sign we needed to say that we would now move Apple from equal weight to underweight.

What stands out is that the AI saw a weaker signal for Apple even at all time highs.

Alphabet: twelve days of room

The signal peaked at 371.6 on May 1. The stock did not top until May 13, at 402.38, by which point the signal had already fallen 42% from its high, and it collapsed to 5.7 by the end of the month. Predictability slid from 0.43 in March to 0.30 in June, so confidence thinned alongside the reading. That combination is the underweight configuration, and it arrived while Alphabet was still making highs.

The signal never recovered. It averaged 30.2 over the last ten readings in the archive against 115.8 for the twenty before them, and it read 14.2 on July 28, near the floor of its range for the entire period. For nearly three months the algorithm gave one answer on Alphabet and kept giving it.

Earnings confirmed it. On July 22 Alphabet beat on the top line with revenue of $119.8 billion against $116.9 billion expected and Cloud growing 82%, then raised 2026 capital spending guidance to $195 to $205 billion from $180 to $190 billion, against the $186.4 billion analysts had modeled. Quarterly capital expenditure hit $44.9 billion, up 101%, pushing free cash flow to negative $5.9 billion. The stock fell more than 6% the next day and reached $317.69 on July 23, down 21.05% from its May high. 

Microsoft: the algorithm held its position through the drawdown

Microsoft looked like the algorithm’s worst call in this set. The signal spiked to 596.1 on May 31, the stock topped the next day at 460.52, and it then fell 23.38% into June 25, finishing the month down 17%, its steepest monthly decline since December 2000, after guiding capital spending toward $190 billion.

The signal never agreed. Through the entire decline it stayed bullish, never printing below 203.6 and averaging 321.8, while predictability climbed from 0.53 to 0.60, the firmest confidence trend of any ticker here.

Earnings settled it. On July 29 Microsoft reported revenue of $90.01 billion against $87.62 billion expected and adjusted earnings of $4.74 against $4.24, with Azure growing 43% and crossing $100 billion in annual revenue for the first time. Management guided Azure growth to 45% against the 41% the street modeled and cut the calendar 2026 capital spending forecast to roughly $175 billion from $190 billion, removing the exact concern that had broken the stock in June. Shares jumped 8.13% to $422.30, adding roughly $260 billion of market value in a single evening.

The signal spent two months looking wrong on price while reading the business correctly.

The Last Week of July

Three of these positions were tested inside nine days. Alphabet reported on July 22, Microsoft on July 29, Apple on July 30. Each result confirmed what the signal had been saying for weeks or months.

Alphabet had been marked down since the first day of May, twelve days before the stock topped, and the signal sat near the floor of its range through the whole decline. The capital spending guidance that arrived on July 22 sent the stock down more than 6%. Microsoft had carried the opposite reading, holding above 203.6 with rising confidence through a 23% drawdown that made it look like the worst call in the set, and its earnings added roughly $260 billion of market value in a single evening. Apple had been the strongest overweight in the book until the signal began fading in mid July while the stock was still setting records, and it fell 7.35% on the guidance that followed.

Two hyperscalers, the same business model, the same enormous AI budgets, and the same daily file separating them in early May. A manager reading sector weights alone owned both at benchmark and learned the difference in late July. A manager reading the file was positioned for it.

That is what this framework buys. Not a model that picks stocks, and not a replacement for judgment, but a ranked read every morning on the small number of sectors and megacaps that already consume most of an institution’s tracking error budget. The overweight and underweight process is not new, and every firm reading this already runs one. What changes is how early the evidence arrives, and how much of the move is still ahead when the weight gets set.

I Know First delivers this forecast daily to institutional clients, on a standard universe or a custom list of tickers. Details at iknowfirst.com.