Curve 創辦人 Michael Egorov：Yield Basis 如何處理無常損失

0:00破解流動性池的無常損失

0:00大家好，我是 Michael，Curve 和 Yield Basis 的創辦人。今天我要跟大家分享 Yield Basis，這其實是一個運行在特殊 Curve 池上的策略，後來發展成自己的協議。現在這個時代，大家都在尋找真正可擴展的收益，那我們到底有哪些選擇？I'm Michael, founder of Curve and Yield Basis, and I will talk about, actually, Yield Basis, which is essentially a strategy running on top of a special Curve pool turned into its own protocol. So, well, I guess right now it's kind of an institutional time, so everyone is looking for a real scalable yield, and which options do we have?

0:48在加密貨幣領域，要找到可擴展的收益，我不是指靠激勵發放的那種收益，而是真正的收益。我們有哪些選項？最明顯的就是借貸，因為有人支付利息，這確實能產生收益，特別是針對美元計價的資產。不過，如果你拿 WBTC 去借貸，實際收益其實很低，所以很多人寧願直接持有 BTC，不想放到借貸平台。For scalable yield on crypto. And I'm not talking about yield based on incentives. I'm talking about, like, real yield. So what options do we have? We obviously have lending, which produces yield by someone paying interest. And it actually works, especially for USD-denominated things. But when you lend out, I don't know, WBTC, for example, the real yield in lending for that is fairly small. So that's kind of not great. So maybe people would rather hold BTC in its pure form without actually lending platforms, right?

1:40還有一種叫做基差交易（basis trades），這在美元市場蠻有趣的，像 Ethena 就有做。不過在加密貨幣市場，基差交易效果不佳，因為大家都太看多加密貨幣了，沒辦法像傳統市場那樣賺到基差收益。你也可以做 covered call 或選擇權，這些雖然有效，但會限制你的上漲空間，也不是純加密貨幣計價的收益，所以也不是很理想。So there are also basis trades, which are fairly interesting for USD. Ethena is doing that for USD, but for crypto, basis trades don't quite work because everyone is—even now, everyone is seemingly too bullish on crypto, so you cannot really earn on basis trades on crypto. Well, you also can do, like, covered calls and options and stuff, and that obviously works, but it's not really crypto denominated. It's something which caps you upside. So also not very good to earn yield on crypto.

2:20AMM 與無常損失的本質

2:20最直覺的方式，當然就是把加密貨幣放到自動造市商（AMM）裡賺手續費。但這就會遇到無常損失（impermanent loss）這個老問題，這一直困擾著 AMM。作為 Curve 的創辦人，我非常了解這點。Curve 之所以受歡迎，就是因為大多數用途是穩定幣對穩定幣，這種情境基本不會有無常損失。Well, obviously, a natural choice would be to earn on crypto by providing it in automatic market makers. But you have this problem of impermanent loss, which was always haunting automatic market makers. And as Curve founder, I definitely agree on that. Like, why Curve is... is essentially, is fairly popular is because most of the Curve use case is stablecoin to stablecoin market making, which is not really affected by impermanent loss.

2:59那如果是加密貨幣對穩定幣呢？這時候大家最愛的其實是質押，像機構都很喜歡質押型 ETF，賺 ETH 或 SOL 的收益，這確實有效，但只適用於已經質押的資產。那有沒有辦法賺更多？其實，解決無常損失是有可能的，這也是我用 Yield Basis 做到的。But what about crypto to stablecoin, right? And of course, something which does work. And is loved by all the, like, institutions and everything is staking. So there are staking ETFs which earn yield on ETH, Ether, or maybe on SOL, and that obviously works. But, you know, that's only available for staked assets and—Well, maybe you can actually make more there, right? And apparently, it appears that it's possible to solve impermanent loss, and that's what I did with Yield Basis.

3:43Yield Basis 已經運行一年了，我想跟大家分享這一年來的資料，看看在這段從牛市高點到熊市低點的市場週期裡，實際表現如何。首先，來說說 Yield Basis 的運作原理。And Yield Basis has been running for about one year, so I wanted to review one year of data, how it actually performed in... Well, in this time of the market, which went from essentially top to hopefully bottom, right? So let's see. Well, first of all, how yield basis works and why, right?

4:10數學模型與槓桿解法

4:10假設我們有最簡單的流動性池，比如 Uniswap v2 這種常數乘積池，它的定價函數是非線性的（開根號）。如果你把同樣的資產放在 AMM 裡，跟什麼都不做直接放著比，撇除手續費收入，AMM 的非線性回報會讓你輸給單純持有，這就是無常損失。So if you look at the simplest possible liquidity pool, let's say, I don't know. A constant product pool like Uniswap 2, right? Its pricing function is very nonlinear. It's a square root function. If you take exactly the same assets and instead of putting them in an AMM, put them on a shelf and let them do nothing, then if you disregard the fees earned, the assets do in nothing will actually do better than your nonlinear payoff function of AMM, which is what we call impermanent loss.

4:58那能不能消除這個問題？答案是可以。如果你有一個開根號的回報曲線，想讓它變線性，你就要把它平方。在 DeFi 裡怎麼做？就是加槓桿，而且是複利槓桿。舉例來說，你有 BTC/USDT 的流動性，把一半的 USDT 拿去借，再買更多流動性，讓債務和流動性維持 2 倍的比例。這樣一來，價格曲線就被平方，開根號消失，還能賺到手續費，看起來就像魔法一樣解決了無常損失。So, is it possible to get rid of it? Apparently, the answer is yes. So. If you have something like a square root, you want to get rid of it. What do you do? How do you make it linear? Well, I guess you want to square it. And what do you do to square it in DeFi? Apparently you can do it. You apply leverage. So for leverage, and it's compounding leverage. So let's say you... Have your liquidity, BTC to USD in this example, and you borrow USD against, like, half value of this liquidity in USD and buy more of this liquidity to keep this ratio of, like, debt versus liquidity being always around factor of two. Well, you do that, then you essentially have the price squared, and square root disappears, and you have fees earned. So it looks like magically solves IL.

6:02問題是，這真的能解決無常損失嗎？看起來是可以，但回報部分就比較複雜了，因為你要維持槓桿，會有成本。你必須用借來的錢買 LP，或賣 LP 還債，這些都會產生成本。你必須賺到的收益要大於這些成本才行。And the question is, does it actually solve IL? Apparently it does, but the returns part is something which is tricky because you do spend something to keep this leverage constant, right? Because you need to buy LP with what you borrowed or sell LP to repay some of what you borrowed. So that is, that costs, right? And you essentially need to earn more than you lose.

6:41集中流動性與策略調整

6:41如果你用 Uniswap v2 這種傳統 AMM，這招其實沒什麼用，因為你賺到的收益大概等於你為維持槓桿所付出的成本。但如果你用集中流動性（concentrated liquidity），情況就不一樣了。Curve 從 2021 年開始就有自動管理的加密貨幣池，這些池子的流動性會根據價格自動調整，但只有在有儲備時才能移動。如果儲備用完，流動性就會卡住，直到賺到更多收益再移動。And apparently, if you have AMM like Uniswap 2, that wouldn't work because you would earn about exactly as much as you lose to keep your leverage constant. But if you have some concentrated liquidity, that already helps. Coincidentally, in Curve, since 2021, we have crypto pools which do have concentrated liquidity, which is automatically managed. But it's not always moved, like, to the current price. This concentrated liquidity follows the price only when there are, you know, there is a reserve to move it, right? So if this reserve is empty, the concentrated liquidity is stuck until—More returns are earned, and then they are spent to move it, right?

7:35所以，是否有足夠的儲備取決於這些加密池的參數。我們在 Yield Basis 做了一些調整，應用了這種「再槓桿」方法來消除無常損失。這在模擬中效果很好。So anyway, whether you will have enough reserves depends on parameters of these crypto pools, and yeah. So we did some modifications for that in Yield Basis, and basically applied this. How we call it, re-leverage method to get rid of impermanent loss. So, that's how it looked like in simulations, right?

8:04模擬與實際資料驗證

8:04在模擬中，你可以看到比特幣價格在兩年多的時間裡，收益曲線有時穩定上升，有時持平再上升，整體看起來很不錯。但實際運作呢？現在我們有一年的真實資料可以檢驗。So, you have on the top, you have Bitcoin price over, how long, over two years, I guess, two years and a half, and how returns would go over that time. So, well, sometimes they are going steady up, sometimes they go flat and then steady up again. So, looks pretty good. So, question is, does it work like that in practice? Well, now we have one year of real data to see how it went.

8:42簡單來說，Yield Basis 的收益來源主要是換匯手續費，但也有支出，包括移動集中流動性的成本，以及維持槓桿的成本。如果扣掉這些支出還有剩，那就是利潤。這些利潤會分給流動性提供者和 YB DAO，鎖倉 YB 代幣的使用者可以分到系統賺到的手續費，作為交換，YB 會有部分代幣發放給流動性提供者。這個分配比例由 LP 自己決定，他們可以選擇要賺手續費還是 YB 代幣。And just to, well. Just to summarize where the returns can go, where the returns go in yield basis. So we have income into the system coming as swap fees, right? And you do have expenses. Some of the expenses go to repay moving concentrated liquidity, right? So moving concentrated liquidity in Curve pools costs something, and another expense is keeping leverage constant. That also costs something. If you have something left, this is profit. Where does the profit go? So the profit is split between liquidity providers who have, you know, this growth. And YB DAO. So locked YB tokens get the fees which the system earns, right? And in exchange, because it takes some of the returns away, YB essentially has some token emissions which go to liquidity providers. And this split between how much is going in token emissions and how much is going in fees is chosen by liquidity providers themselves. So they can choose whether they earn fees or whether they earn YB tokens, and how much YB DAO gets is also dependent on that. So, but when we analyze data, we don't want to look at this split. We want to actually see how much the system earned as a whole. So, like, essentially how much profit was left after we used the expenses.

10:33策略調整與風險管理

10:33我們一開始用的是比較激進的池子參數，收益非常好。你可以看到模擬的虛線是保守參數，而我們的綠色實線一開始漲得很快。不過，激進參數會讓系統和穩定幣承受很大壓力，必須調整得更保守。太激進的話，收益曲線會變得太平，甚至可能出現損失。所以我們後來改成更保守的參數，綠線繼續上升，整體表現跟模擬差不多。我們也找到在安全性和收益之間的平衡點。And here is what we've got, right? So we started with... Fairly aggressive pool parameters, which earn really, really well. So you see this dashed line showing a simulation, right? But this simulation is at very conservative parameters, and we started at fairly aggressive parameters. So you see the green line going up very, very well. But, you know, it's aggressive. And what does that mean? It means that sometimes the system and, like, the stablecoin will experience a very serious stress, and we need to run more conservatively. And another kind of another consequence of that is this yield kind of flattening too much. Because the parameters were too aggressive, so we, and yeah, you can have some losses as well when you have aggressive parameters. So we changed to more conservative parameters, and that's where you essentially see the green line continuing. But overall, overall, we are fairly close to what simulation shows, and we figured out how to, well, how to find balance between good, like, safety of the system and earnings, right?

12:06但這個工作還沒完全結束。我還想再加兩個改進，一個是更聰明的動態手續費，當市場波動變大時，手續費也要提高。我們現在有做一些，但還可以更改善，而且這不需要流動性遷移。另一個更有趣的是類似 prop AMM 的動態，讓外部交易者把價格資訊帶進系統，這樣系統可以利用這些資訊賺更多。模擬顯示這很有效，而我們也看到模擬和現實相符，這部分很快就會上線。But this work is not, is not really, not really fully complete. We... I actually want to apply two more things to, I don't know, to make the system safer and at the same time make it earn more. So one of the things is doing smarter dynamic fees and, well, basically if volatility becomes higher, you want fees to be bigger, right? We do some of that. But it's probably not as optimized as it can be. So we are thinking deeply about that, and we can change it without actually any liquidity migration. But another probably more interesting is dynamics very similar to prop AMMs, where external traders, they actually bring price information into the system. And system utilizes this price information and apparently earns more. It's fairly easy to see in simulations, and as we saw, simulations do match reality. So we are configuring that part as well, and it will be live actually very, very soon.

13:22總結：無常損失已被破解

13:22總之，我們已經打造出這套系統，經過實戰考驗，還有更多改進即將推出。你現在可以說，無常損失這個問題，已經被解決了。謝謝大家！Anyway, so we have this system made. Battle tested, and there are improvements going where you essentially have impermanent loss solved. That's it.