The semiconductor space has long been viewed as one of the best leading indicators of the modern technological economy (and capital markets), and with good reason: it was the first sector to bottom in March 2020 when it became apparent that China and various western central banks would inject trillions into the global economy, and had enjoyed a nearly relentless upward climb since then peaking just over a week ago on August 4, but then something snapped...
... and as shown in the next chart, the Semi Index is broadly lower again, having dropped for six consecutive sessions - the longest such streak since the October 2018 Fed "policy error" when stocks cracked after Powell threatened to tighten far more than markets expected, only to end his hiking cycle prematurely just two months later, resulting in the first bear market in a decade.
And while Pavlovian dip-buyers have emerged to bid every daily drop in the SOX - after all it will take a major reversal to undo 15 months of upward momentum...
... a sustained reversal in this critical market-leading indicator would have profound consequences for overall risk tone and could very well signify a market top if it fails to reverse higher.
To be sure, so far the pain has been scattered, hammering mostly chip giant Micron Technology which fell as much as 8% amid concerns about memory demand, while peers Western Digital slid 7.8% and Seagate Technology fell 3.8% despite an upgrade from Goldman Sachs. Intel also fell, sliding 1% and Texas Instruments dropped a modest 0.7%.
But if Morgan Stanley is right, the pain for semis - which have long been a proxy for outsized China's monetary generosity and spending - is just starting because, in the words of the bank's equity strategists, "Winter is coming."
In the report which downgrades the DRAM space in general, and several chip giants such as Micron and SK Hynix in particular, Morgan Stanley warns that "while pricing is still moving higher, the rate of change is approaching peak as supply is catching up to demand."
Adding to this, the bank's cycle indicator has shifted out of "midcycle" to "late-cycle" for the first time since 2019 and this phase-change has historically meant a challenging backdrop for forward returns.
Asking rhetorically "What's changed", MS gives the following response:
Cyclical conditions for DRAM have started to roll over. Initially, we believed that as long as demand remained strong into 4Q, prices could continue to go up since bit supply has outpaced production and help tighten inventory. However, while demand has stayed strong on a relative basis, it has worsened in recent weeks which has led to a downtick in pricing expectations. Initial indications are pointing to a more challenging environment for pricing into 4Q following prolonged 3Q negotiations that resulted in continued price hikes, and a reversal in trend into 2022. What this means in practice is that the investment case for memory will migrate from "what is the earnings potential?" toward "why should I own DRAM at this late stage, and what price is fair?". We are trimming estimates to reflect worsening DRAM conditions assuming an earlier peak (3Q21) and high single digit quarterly declines in 1H22.
As for why MS sees winter coming sooner, it says the following:
A cyclical downturn is likely from early 2022... Playing the second derivative inflection in DRAM (measured in terms of the monthly rate of change in contract pricing illustrated in Exhibit 6 ) could see contract pricing fall a long way – typically -50% YoY for both blended and spot prices – and we forecast material downside for memory prices and earnings. Memory valuations have come down since their 1Q peak but earnings expectations have not. Both supply and inventory present higher downside risk as we move to 1Q22, despite structural growth in cloud that continues. The current memory cycle is currently in a delicate phase with its own version of inflation (price) vs. growth scare (component constraints). As we move closer to 4Q21, the main investor debate at present will be around the durability of cyclical positioning against a backdrop of peaking prices.
… and inventory driven, given limited expansion in new capacity and underlying supply growth. The duration of the down cycles has more or less remained the same, but the magnitude of the sales decline may not be as large as before. In our view, this is due to the structural changes that have taken place in the DRAM industry – consolidation, a traditional barrier to entry – has increased and demand drivers are broadening and we expect the trend of softer down-cycles (higher lows) to continue to be the norm going forward. What were changes in PC demand as the dominant end-use market in the past, has now shifted to the number of end-use products that continues to grow, such as AI-enabled chips (in mobile, server and automotive industries) in need of larger die sizes (DDR5), and applications requiring high memory band width. This permeation into virtually every sector of the broader economy has made the industry more dynamic by diversifying its long-term growth prospects and likely resulting in milder down-cycles
The bank expects that once earnings growth expectations reverse, the space could suffer a near-30% PB valuation contraction and higher chances of positioning reset. Actually, the contraction could be far worse with the bank noting that "downcycle transitions commonly see P/B valuations contract by 50-64%. DRAM stocks can pull back 32-63% from peak to trough at any given downturn – we are now 12% lower from the April 2021 high – and the average stock price was down 43% in the last two down turns. The risk of disappointment is high in the late stages of the cycle as price hikes have historically been a 'sell the news' events. While the risk premium compresses following excitement around recovery, eventually positioning resets, and a slowing rate of change in growth offset this and sent multiples lower. Structurally higher earnings and margins through the cycle mitigates downside, but the pricing and inventory correction that follows limits the upside. The net effect is a tug-of-war between earnings/valuation, tepid returns over the next 12 months, and a likely 30-40% correction in P/B valuations between here and there."
Amid these inflection signposts, the bank warns that "sell signals are accumulating" and notes that while it was "prepared to become more constructive under the right conditions", ultimately it is about inflections in the cycle and trajectory of earnings estimate revision breadth – the former approaching an earlier peak YoY pricing and latter approaching negative earnings risk that follows."
Here is the bank's core thesis:
- the next cyclical downturn begins from 1Q22 and DRAM will stay fundamentally oversupplied in 2022, exacerbated by inventory builds;
- recent growth indicators have downticked, while demand held back by component supply;
- valuation is no longer compelling on the way down; and tactically buying on the dip no longer makes sense from a quant perspective – better entry points will come in the future.
There are other challenges facing the DRAM industry including higher capital intensity and a flattening cost curve, which has created a fundamental change in chipmakers’ capital allocation decisions and return requirements. This is evident with DRAM industry operating margins troughing at a low-30% level in 2019, vs. losses in previous downturns and capital expenditure discipline via capex cuts, underpinning a steady rise in return on capital.
The box below summarizes the key considerations that prompted the downgrade:
Morgan Stanley also has a remarkable chart showing the multiple expansion spread between EPS and underlying prices, a delta which Morgan Stanley attributes to "global easings" and which would "be gradually narrowed as time goes by." Some more commentary:
Risk sentiment is turning to risk-off with elevated valuation pressure amid rising UST 10Y yield: The low rate and global easing environment have significantly lifted the valuation of both market and tech stocks. We flagged Exhibit 47 in May to remind investors that majority return of semiconductor stocks was expectation driven, and contributed by multiplier expansion. Although the improved industry fundamentals met market expectations in 1H2021, the valuation premium that the market had previously priced in should be normalized over time either by further pick-up in EPS or an absolute market correction We see the current market dynamics is developing toward the latter scenario that Fed's tapering is coming soon, while US Treasury 10Y yield tended to rise than decline in 2H2021 amid high inflation. On the other hand, investors are taking risk off from their portfolio YTD (ex). These effects in combined would have net negative implications on the valuation on semiconductor industry. Given memory players are subject to higher risk of earnings momentum slowdown, we think the embedded downside risk is emerging as we move into late-cycle of DRAM price.
With this in mind, Morgan Stanley prefers NAND/NOR over DRAM and has downgraded Hynix to UW, while also downgrading Micron, Nanya Tech and Winbond to EW, and shares the following aphorism:
History says that stocks will tell us when it's time to buy, once they stop declining in reaction to bad news – but the bad news has to come first, and our downgrades partly reflect this process.
On the industry side, MS is now an average 10% below consensus on 2022e EPS and see stocks moving back below mid-cycle multiples which form the basis of its PTs. .
The full MS report can be found in the usual place for Pro subs.