渠乾利贞 止损持盈

Secondary market craftsman. Cycle player.
No shilling. No calls. No referral fees. Just journaling.
PnL is your own. Not financial advice.
Don't follow. Don't tip. Read and move on.

Doing the secondary level. Only work in cycles.
No shouting buy/sell signals, no carrying trades, no doing rebates, no taking advertisements.
Writing is for myself, not for you.
Profit and loss are your own responsibility. This does not constitute investment advice. Don’t pay attention, don’t tip; if you’ve seen it, you’ve seen it.

—— Clear Stream Channel

21Shares has filed a revised S-1/A for its spot Injective ETF, code TINJ, on Nasdaq.

On September 19, the SEC received 21Shares’ amended filing. This is an updated version of the original application submitted in October 2025. If approved, TINJ will be listed on Nasdaq, passively tracking the FTSE Injective Index, and it can also use part of its holdings to stake and earn yield.

21Shares has already been doing Injective staking ETPs in Europe for a while. This time, it’s bringing essentially the same setup to the United States.

Pay attention to this detail: staking.

This is the most interesting part of this latest wave of ETF applications. In 21Shares’ ETF application for Hyperliquid, it mentions staking 30% to 70% of the holdings. The Injective filing also includes a staking option.

What does staking mean? It means the ETF isn’t just passively holding tokens. It locks up the tokens, earns network rewards, and then counts that portion of the rewards as part of the fund’s return.

In the SEC’s eyes, this has been controversial all along. Does staking count as issuing securities? Do staking rewards count as consideration under an investment contract? Those questions have previously blocked the staking versions of ETH ETFs. Now 21Shares is trying first on smaller assets like INJ and HYPE.

But don’t confuse an application with approval.

21Shares has a long queue of S-1 filings. Solana, SUI, SEI, ONDO, and also Hyperliquid. SUI is already listed on Nasdaq under the ticker TSUI. The Injective filing moving to S-1/A means the SEC has issued its first round of comments and 21Shares is responding. There’s still a ways to go before the 19b-4 approval to list.

And Injective isn’t exclusive to 21Shares. Canary’s Staked INJ ETF was submitted to the Cboe back in July 2025. 21Shares has experience with Injective ETPs in Europe, but in the U.S. market, first-mover advantage may not necessarily belong to it.

— Qingliuqu #sec收到21sharesinj现货etf修订申请

S&P Dow Jones announced on September 4 that the changes would take effect before the start of trading in U.S. stocks on September 21. Sandisk was upgraded from the S&P 500 to the S&P 100. At the same time, the companies kicked out were Colgate, Nike, Honeywell’s aviation unit, and Simon Property Group. The four new entrants were Dell, Palo Alto Networks, Arista Networks, and Sandisk—all tech stocks.

Pay attention to one detail

Nike had been in the S&P 100 for 18 years, and this was the first time it was removed. Over the past five years, the S&P 100 rose 83%, while Nike fell 80%, with a market cap loss of $230 billion. It’s not that Nike did something wrong—it’s that money is flowing into AI, and Nike isn’t on that track.

Sandisk is running on that track. And it’s running faster than anyone else.

In February 2025, it was spun out of Western Digital and went public at $35 per share. By the close on September 18, it was $1,791. In a year and a half, it jumped more than 4,300%. Over the past 12 months, it rose 1,700%.

But the most ruthless part isn’t the surge. It’s its financials.

In fiscal year 2026’s fourth quarter, revenue was $8.97 billion, up 372% year over year and up 51% quarter over quarter. Full-year revenue was $20.25 billion, up 175%. Its data center business grew 437%.

Gross margin is 56%, ROE is 91.6%, and ROIC is 102%.

Have you ever seen a hardware company with this kind of profit margin? This isn’t selling flash memory—it’s selling a money printer.

The news that Sandisk would enter the S&P 100 was released on September 4. But before September 4, it had already had a run. On the day the news came out, it jumped 11.9% and closed at 1,740. Then it pulled back to 1,519, before rallying back to 1,791. On September 18, it rose another 11%.

My take

Sandisk is one of the purest plays in this AI infrastructure rally. It’s not chip design, not computing power—it’s storage. AI training and inference require massive amounts of NAND flash memory, and Sandisk is one of the largest suppliers. Its backlog is $93.9 billion, of which $16.5 billion is secured.

But look at the valuation here now. 1,791 is 52-week high of 2,354—down 24% from the peak. The forward P/E is 8.37x. For a company with revenue up 372%, it’s cheap to the point of being unbelievable.

— Qingliu Channel #闪迪将于9月21日纳入标普100

BTC once again surged above 80,000—feels like it’s finally holding strong.

Short covering is the fuel, and spot buying is the engine. Once the fuel runs out, it’s gone—but the engine can keep running.

Things are changing on the macro side too.

The U.S. Treasury Secretary rolled out Treasury repurchase operations. On the surface it’s a technical move, but in reality it’s like putting a steadying balm into the market. The market has started trading again under the logic of “improving liquidity” and “a weaker dollar.” Gold and Bitcoin benefit at the same time—this isn’t a coincidence. When sovereign credit is questioned, scarce assets get repriced.

On top of that, the SEC has introduced a new regulatory exemption framework, giving tokenized securities trading a five-year green light. With regulatory uncertainty decreasing, the psychological threshold for institutions to enter the market is also lowering.

There’s a wall above 80,000. Large sell orders are clustered in that zone—a liquidity wall. Whether it can be absorbed depends on whether spot buying can keep going.

The HKMA has issued a release: by the end of the year, it will roll out a wholesale central bank digital currency (wCBDC) for interbank tokenized-deposit settlement, enabling 24/7 payments. It is called Project EnsembleTX, and a live pilot was already launched toward the end of last year.

Many people see the three letters “CBDC” and get excited, thinking Hong Kong is about to make some big move. Others see “CBDC” and panic, believing the government is coming to monitor every cent you have.

Both reactions are wrong.

First, let’s clarify what this wCBDC is.

Wholesale, not retail. That means it’s not for individual use—it’s for banks only. It solves a very specific technical problem: Hong Kong’s cross-bank settlement currently relies on the RTGS system (Real-Time Gross Settlement), i.e., real-time, full-value settlement. This system operates only during business hours. If you want to transfer a tokenized deposit at night, sorry—you have to wait until the next morning.

What the wCBDC does is to move that settlement step onto a blockchain so it can run 7×24 hours. It can also support margin settlement for post-market derivatives trading at the HKEX.

In plain terms, this is an upgrade to financial infrastructure, not a currency revolution.

Why is Hong Kong doing this?

Look at the use cases it chose—you’ll see immediately. Tokenized deposits, tokenized government bonds, and margin settlement for HKEX derivatives. All are institutional scenarios. All are the parts in traditional finance where “settlement is too slow and costs are too high.”

HSBC, Bank of China (Hong Kong), and Standard Chartered are all on the pilot list. They’re doing the same thing: using digital HKD to pay margin for post-market derivatives trading, so settlement is no longer constrained by bank closing hours.

The HKMA itself is very clear: it will prioritize developing wholesale use cases, and the launch of digital HKD for retail use “has not yet been decided, or when.”

So what does that have to do with the crypto world?

Honestly, not much in the short term.

wCBDC is not a stablecoin. It won’t be listed on exchanges, and it won’t go into DeFi. It is a central bank liability—a settlement instrument for interbank transactions—running on a permissioned blockchain. Your USDT and USDC are in a completely different category.

You could even say it competes with USDC. If interbank tokenized-deposit settlement can be handled with wCBDC, then stablecoins’ room to operate in institutional settlement scenarios would be squeezed.

If it gets going, Hong Kong could become a settlement center for tokenized assets in Asia #香港拟年底前推出批发cbdc

Bottomline, one of the world’s top three SWIFT service providers, processes 160 trillion in payments per year and has 600+ banking clients. It built a platform called Global Pay Connect that enables these banks to connect directly to on-chain settlement using ISO 20022 standard messages, with Chainlink’s CCIP and CRE.
Think about that design.
Banks don’t have to replace their systems, buy crypto, or learn something new. They just send the original payment instructions; Chainlink translates them into on-chain operations behind the scenes, and then settles.
This is bigger than it looks on the surface.
Bottomline handles about 15% of SWIFT cross-border transactions globally. This isn’t a fringe player. It’s an active artery in traditional finance infrastructure. Now that artery has a blockchain branch connected to it.
And it didn’t choose to build its own chain. It chose Chainlink as the middleware layer. CCIP handles cross-chain messaging, while CRE orchestrates the payment workflow. Banks only need to connect once to reach multiple chains.
But don’t equate “integration” with “usage.”
Neither company has disclosed how many banks will actually use the feature, published rollout timelines, or shared transaction volumes. Integration is a technical capability; usage is a business decision. There’s a world of difference between the two.
Bottomline’s own product lead once said: whether to adopt it depends on whether the finance team can manage on-chain payments with the same level of visibility, control, and governance capability as it uses to manage existing payment methods. Translated plainly: the tech works, but the finance department may not buy in.
Chainlink’s standing is changing.
In the past few months, it won a bundle of institutional deals. Aave set CCIP as the default cross-chain infrastructure. BitGo chose CCIP as the exclusive cross-chain provider for WBTC, managing $7.3 billion in assets. Coinbase picked it as the oracle for tokenized stocks on Base. Nethermind migrated validators from LayerZero to Chainlink.
This time, it’s the SWIFT ecosystem. Most cross-border payment messages worldwide originate from the SWIFT ecosystem. Bottomline has given CCIP a direct channel into that ecosystem #bottomline推出chainlink链上支付平台

Circle spent more than a year building an institutional-grade setup. The list of founding verifiers includes BlackRock, Visa, Mastercard, DTCC, and ICE. The testnet processed over 700 million transactions. The whitepaper is all about tokenized funds, stablecoin FX, institutional settlement, and AI agent economics. CEO Allaire’s exact words were: “the economic operating system of the internet.”
Then on day one after launch, in came the meme-coin crowd. 97,025 new tokens were minted, 83,751 of them from Arguspad.
I’m not mocking Circle. I’m talking about something more real.
For any new chain, the traffic on day one has never come from “real use cases.” It comes from speculation. Speculation doesn’t need education, doesn’t need compliance, doesn’t need institutional approvals. All speculation needs is one thing: someone willing to bet.
Arc’s USDC gas, sub-second finality, and EVM compatibility—those technical features are selling points to institutions. But to meme players, they only mean one thing: fast, cheap, and able to mint tokens.
So Arguspad ate up half the chain. This isn’t Arc failing—it’s the normal pattern for every public chain at launch. On day one, the Robinhood Chain had only 568,000 in volume, and it took more than a week to slowly pick up. Arc did 410 million on day one because Aave, Uniswap, and Morpho were already deployed in the very first block—liquidity was in place, and the launchpad was directly getting piped in.
What’s really worth looking at is the 75 million after you remove the launchpad.
That number is what corresponds to what Circle truly wanted to do: USDC settlement, tokenized collateral, StableFX FX, and institutional payment rails. For a new L1, put next to 336 million of speculative volume, the ratio is 1 to 4.5.
This is the dilemma facing all “institutional chains” right now. You set up the stage, lay down the red carpet, and send invitations to BlackRock and Visa. Then the first people to rush in are the ones in flip-flops trying to grab the airdrop.
I’m not judging whether that’s good or bad.
In September 2026, on the launch day of a new chain, speculation is still the only thing that can instantly fill the blocks. Institutional narratives still need compliance processes to run their course. Meme coins only need a contract address.
Circle is betting on the long term. But the market never gives votes for the long term—it only votes for today. #meme发射台占arc首日成交82%

On Solana, most USDC is used for payments, transfers, and DeFi interactions—it’s “spent money.” On Hyperliquid, almost all USDC is margin for perpetual contracts—it’s “staked money.” The former is water flowing through pipes. The latter is bullets waiting in the chamber.
Think about the difference in that picture.
Why can Hyperliquid be so strong?
I checked the data: in the past 30 days, Hyperliquid’s perpetual contract trading volume was $240 billion, the highest. The second-place Arbitrum had $47.2 billion, while Solana had $46 billion. Do the math on that multiple—it’s over five times.
With trading volume stacked there, margin naturally piles up there too.
Plus, Hyperliquid’s USDC has a special feature. This past May, Circle, Coinbase, and Hyperliquid did a partnership. USDC became Hyperliquid’s only quoted asset. Coinbase manages the treasury deployments, and Circle handles minting and redemption. Even more importantly, 90% of reserve earnings are sent back to buy back HYPE.
Take a closer look at this design.
Users deposit USDC as margin. The interest generated by that USDC doesn’t go to Circle or Hyperliquid. Instead, it’s used to buy HYPE in the market and then burn it. For every additional dollar of USDC, there’s an additional amount of potential pressure to buy HYPE.
This isn’t just a stablecoin. It’s a flywheel.
But what I want to talk about today isn’t that.
I want to say that what this data truly reveals is something deeper: the “use cases” of stablecoins are splitting.
Previously, everyone looked at stablecoins in terms of total supply. How many billions of USDT, how many billions of USDC—who’s growing and who’s falling. But now, if you look at on-chain distribution, you’ll see a trend: stablecoins are starting to branch out by “function.”
On Ethereum, USDC is the vault for old money. On Solana, USDC is the payment channel. On Hyperliquid, USDC is the chips at the poker table.
These three things all get called USDC, but their “turnover rate” is completely different. Money in the payment channel moves many times a day. Money at the poker table moves more frequently too, but in a more one-directional way—it only flows in, not out, until it gets liquidated or withdrawn.
Hyperliquid’s USDC being able to outperform Solana shows that, at some point, the volume of “betting” in this market outweighed the volume of “use” #hyperliquid上usdc供应量超越solana

PMatt Huang yesterday posted on X. It was very simple—just two sentences: Paradigm is an investor in the Zcash Open Development Lab, and it also holds ZEC tokens. He called Zcash “a privacy add-on to Bitcoin.”

Then ZEC surged—up nearly 20% in 24 hours, and up 160% over the past month. How much did Bitcoin rise over the same period? 18.2%.

First, let’s talk about Matt Huang.

He’s a co-founder of Paradigm, one of the top-tier crypto VCs in the space. He never posts casually. You can look up what narrative he publicly backed last time. When someone like him posts, it’s not to hype a trade—it’s to set the tone.

In his post, there are three points that I think are worth watching even more than “buying ZEC” itself.

First, he said Paradigm supports Zcash’s “inflation-funded developer fund.” What does that mean? It means that when miners mine coins, a portion of them will be used to support the development team. In crypto circles, this is a highly controversial mechanism, but he says it’s important because “AI-driven network attack capabilities and quantum computing are developing.”

Second, he supports combining Zcash’s token voting with other forms of governance, reducing its unpredictability as a monetary asset.

Third—and most interesting—he defines Zcash as “a privacy add-on to Bitcoin,” not a replacement. This positioning is spot-on. Bitcoin is a transparent vault, while Zcash is an invisible vault. The two don’t conflict—they’re complementary.

But you need to see one thing clearly.

Paradigm’s investment in the Zcash ecosystem didn’t start yesterday. Back in March, the Zcash Open Development Lab completed a seed round of over $25 million. Investors included Paradigm, a16z crypto, Coinbase Ventures, and Winklevoss Capital.

On September 14th, the Zcash community vote ended. 98.9% of token holders supported maintaining the Bitcoin-style halving mechanism, not opting for smooth issuance.

This was one of the most important governance votes in Zcash’s history. With nearly unanimous consensus, the community confirmed that Zcash will follow the path of Bitcoin’s monetary attributes plus optional privacy.

Then Grayscale’s Zcash ETF, ZCSH, was listed on the NYSE Arca on August 25th. In two weeks, assets under management surpassed $500 million, with holdings of more than 550,000 ZEC—locking up 3% of the network’s total circulating supply.

——Clear Channel #paradigm披露持有zec

September 16, in the U.S. House of Representatives, the Committee on Financial Services advanced the “2026 American Reserve Modernization Act” with a vote of 28 to 21—H.R. 8957. In simple terms, Trump’s strategic Bitcoin reserve has been upgraded from an executive order to a federal law.

First, what exactly is this bill meant to do?

In one sentence: Lock up the U.S. government’s 324,000 bitcoins for at least 20 years so they can’t be sold. The value is about $24.77 billion.

They can’t be sold, exchanged, auctioned, or pledged as collateral. The Treasury Department must establish secure storage facilities, regularly receive independent third-party audits, and issue reserve-proofs reports each year.

It sounds pretty tough, right? But then you look one layer deeper.

The original version was called the BITCOIN Act, aiming to buy 1 million bitcoins within five years using a budget-neutral strategy. Now, in the ARMA version, they’ve removed the purchase target. What did they change? They take the bitcoins the government already has—seized from criminals—and put them into the reserve.

Think about the difference.

In the original version, “we actively go buy.” In the new version, “we lock up what we already have.”

One is offense, the other is defense.

Why did it become defense? Because of political reality.

This bill has 23 co-sponsors: 22 Republicans, and the only Democratic co-sponsor, Jared Golden, isn’t even a member of the Financial Services Committee. Within the committee, there are 23 Democratic lawmakers—none of them signed on.

This isn’t cross-party consensus; it’s a partisan line. Republicans have a 30–23 majority on the committee, so they can pass it without any Democratic votes. But what happens when it gets to the full House? And the Senate?

There isn’t even a corresponding bill in the Senate. The House members left Washington after September 17 and don’t return until after the November midterm elections. The only realistic pathway is to attach it to the year-end National Defense Authorization Act.

So why push it anyway?

A few years ago, what was the U.S. government doing? Suing exchanges, pursuing penalties against project teams, treating cryptocurrency as a money-laundering tool to crack down on. Now? The gap between then and now—prices rising from 3,000 to 78,000—is huge.

But don’t confuse “a show of posture” with a “buying signal.” Those are two different things.

If this bill really could pass, that would be a long-term, structural positive. The problem is: it probably won’t pass. And the market is currently pricing in the assumption that “it might pass.”

—— clean stream channel #美众院推进比特币储备法案

We saw the data from Galaxy Research: on September 11, Bitcoin experienced a single-block reorganization at block height 966,500. Spiderpool and Antpool both found two valid blocks at almost the same time. In the end, because Antpool’s chain had more accumulated work, the network accepted it, while Spiderpool’s block was discarded.

This is the third time within four weeks. The previous two occurred on August 16 at 962,722 and on August 24 at 963,853.

What is a single-block reorganization?

Simply put, two mining pools mined blocks at nearly the same height at the same time. The network briefly forked into two chains. Then the next block was added on top of whichever chain came first. That chain becomes the main chain, and the other is discarded. The discarded one is called an orphan block. If the transactions inside it have not yet been confirmed, they go back to the mempool and will be packaged again by the next block.

In Bitcoin’s rules, this is completely normal. It’s not an attack, not a loophole—just two miners accidentally hitting the same time.

The network automatically selects the chain with the most accumulated work. This is the logic written by Nakamoto.

What I want to point out is: why did three of these happen consecutively within four weeks?

Think back to the bull market cycles in 2017 and 2021. When it comes to Bitcoin network reorganizations, getting one once a year would already be considered news. Mining pools’ hash power is distributed; whoever finds a block is an expected event, making the odds of a “collision” extremely low.

The reason is hidden in a metric that most people aren’t paying attention to. Bitcoin’s network hash rate has fallen from a peak of close to 1.3 exahash at the end of last year to around 920 exahash—down by roughly 20%. And for more than 300 straight days, it hasn’t set a new high. This is Bitcoin’s first sustained hash-rate bear market in history.

In the past six months, listed miners reduced their actual hash rate by 15%, shutting down machines totaling 56 EH/s. Cango and IREN alone accounted for more than 60% of the reduction, because renting capacity to AI data centers is more profitable. Core Scientific’s second-quarter gross margin for its own mining business was negative 56%, and the data-center co-location business generated nearly 80 million in gross profit. TeraWulf gets 70% of its revenue from high-performance computing rental. Keel Infrastructure shut down its entire Bitcoin mining business and shifted to AI facility operations. The miners’ holdings index dropped to -1.2 in September, and the amount of newly mined Bitcoin flowing to exchanges has nearly dried up—because they don’t need to sell anymore. The AI side is covering the cash-flow shortfall.

— Clearing Flow Channel #比特币四周内第三次单块重组

The 10-year U.S. Treasury yield surged to 4.979% on Friday—just a breath away from 5%. The 30-year yield is even more ruthless: 5.37%, the highest since 2007.

Think about it: earning 5% while doing nothing risk-free. What does that mean? It means the opportunity cost of buying Bitcoin is 5%. The opportunity cost of buying Nvidia is also 5%. For any risky asset, the opportunity cost is 5%.

Institutions aren’t gambling with their own money. Their money has costs, performance evaluations, and benchmarks. When the risk-free rate reaches 5%, the threshold for allocating to risky assets gets higher. If you’re not rising fast enough, they won’t come. If your volatility is too high, they also won’t come—because for them, they can just earn 5% lying there. Why would they play whack-a-mole with you?

This isn’t just a crypto-market issue—it’s about all risky assets. But crypto reacts the most violently because it’s at the far end of the risk spectrum and the most sensitive asset, with no cash-flow support. When interest rates move, it hurts first.

So why are yields skyrocketing?

Three things are piling on top of each other.

First, oil prices. Brent crude broke through $107, jumping 6.3% in a single day. When energy prices rise, inflation expectations can’t hold down. If inflation can’t be kept down, forget about rate cuts.

Second, PPI. In August, the year-over-year producer price index rose 5.4%, above expectations. Wholesale-side inflation is still climbing; it will take time to transmit to the consumer side.

Third, and most subtly: Trump promised that if Republicans win the midterm election, they’ll send $5,000 checks to all adult Americans. Do the math—roughly $1.2 trillion to $1.3 trillion. Where does the money come from? From borrowing. Borrow more, issue more Treasuries. The market can’t absorb it, so yields keep rising.

What’s most ironic? Treasury Secretary Besen?t (Bessent) hasn’t made no effort. He expanded the size of Treasury buyback operations, trying to push down long-end yields. On Thursday, the operational cap was set at $6 billion—three times the usual routine. But in the end, only $5.19 billion was actually repurchased, not even hitting the cap. For 10- to 20-year Treasuries, buybacks didn’t fully reach the cap—something that’s never happened in history.

Even when the government itself stepped in to buy its own bonds, it still couldn’t push prices up. The market is voting with its feet. This isn’t a problem of insufficient buying demand—it’s a problem of the price being wrong. Unless yields rise to some level, nobody is willing to take the trade.

Just wait. Wait for the September 16 settlement. Wait for the smoke to clear.

— 清流渠 #美国10年期国债收益率逼近5%

Grayscale submitted an S-3/A amendment to the SEC on September 11 to change the Litecoin Trust into a Grayscale Litecoin Trust ETF. The code stays the same—still LTCN. The listing venue moves from the OTC to NYSE Arca.

When the news came out, I looked at Litecoin’s chart and there was little to no reaction. That’s normal. People who understand this already knew. Those who don’t just glance at the headline and move on.

But I think this piece of news is more important than most people think. It’s not important for Litecoin itself—it’s important for the entire market.

On September 9, LTCN’s closing price was $4.01, while the net asset value of the Litecoin assets it holds was $3.39 per share. That means you’re buying the trust at an 8% premium over its actual value.

Why would anyone be willing to pay an 8% premium? Because a closed-end trust has no redemption mechanism. Once you buy it, you’re locked in. The only way out is to sell it to someone else in the secondary market. Whether it trades at a premium or a discount depends entirely on market sentiment.

What Grayscale is doing is opening that loophole. An ETF has a creation/redemption mechanism. Authorized participants can create or redeem shares based on net asset value. Once market makers see any premium/discount, they can arbitrage it, and the price will naturally move back toward NAV.

Look at LTCN’s history to see how important this loophole is. From August 2020 to June 2026, this trust’s maximum premium reached 58.93%, with an average premium of 5.92%. Its maximum discount reached 67%, with an average discount of 2.6%. There were 802 trading days where it traded at a discount.

58.93%—think about that number. An asset’s price can be nearly 60 times higher than its actual value simply because nobody can arbitrage away the spread. This isn’t investing; it’s a closed-door game.

In the eyes of regulators, this is the simplest kind of asset. So Grayscale chose it as the third one. Not because it’s the most valuable, but because it’s the easiest to get through.

Grayscale’s product pipeline: Litecoin is just one stop along a production line. The Dogecoin trust is changing its S-1/A. The AAVE trust is moving to an ETF. HYPE and BNB have registered in Delaware. TAO has been filed. Worldcoin has filed with Nasdaq.

Grayscale is doing one thing: converting all the closed-end trusts it has accumulated over the past decade—one by one—into ETFs. Litecoin isn’t an exception; it’s the third step in a reusable template. The first step was GBTC, the second was ETHE, the third is LTCN, and there’s a long line of others waiting behind.

— Clear Stream Channel #sec收到灰度莱特币信托转etf申请