Security Markets

docs/_toc.yml

@@ -16,6 +16,7 @@ chapters:
 - file: notes/lec10_prediction
 - file: notes/lec11_crypt
 - file: notes/lec12_pos
+- file: notes/lec13_security_markets
 
 
 

docs/notes/lec13_security_markets.md

@@ -0,0 +1,189 @@
+# Security Markets
+
+## High Frequency Trading Arms Race
+
+In 2010, Spread Networks invested \$300 million to create a new
+straight-line optic fiber cable from Chicago to New York, reducing
+latency from 16ms to 13ms. High frequency traders race to make trades
+first, which is why they value the decrease in latency. However in 2011,
+microwaves brought latency down to 10ms.
+
+Is this a market failure? From Spread Network's perspective, the cable
+might have made back its investment (and then some profit), but even
+then the opportunity cost of the investment might have been high as
+well. Furthermore, does high frequency trading in and of itself even
+benefit society?
+
+## Stock Markets
+
+We build on the definitions introduced in Section 11 (Prediction Markets
+and Information Cascades).
+
+```{prf:definition}
+Individuals who want to buy/sell stock as a means to store or liquidize
+assets. Naive investors do not have a primary goal of profiting from
+private information.
+```
+
+With just naive investors, there can be market frictions: there may not
+be stocks to buy or sell when a naive investor wants to buy or sell.
+Furthermore, the no-trade theorem does not apply as these investors
+derive value from the act of buying or selling itself instead of purely
+caring about the monetary value of their assets.
+
+As before, liquidity providers leave resting orders and buy low, sell
+high. They provide an intermediary for naive investors to trade with,
+but naive investors lose the spread that liquidity providers gain. As
+such spread is part of naive investors' transaction costs.
+
+A stock's "true" value represents aggregate beliefs about the stock's
+returns. In equilibrium, a stock's true value is between the stock's bid
+and ask prices (otherwise, some agents have a profitable deviation to
+buy or sell the stock). True value may vary depending on external
+events. For example:
+
+1.  Correlated Stocks/Securities: If someone buys gold in Chicago, the
+    price of gold in New York increases. If there is demand for security
+    $X$ increases and security $X$ is correlated with security $Y$, then
+    the price of $Y$ will go up.
+
+2.  Fed announcements: Fed made announcement on Sept 18, 2013. Markets
+    reacted faster than the speed of light.
+
+3.  Twitter: Trump tweets (about tariffs, the Fed, covfefe) changed
+    stock values.
+
+4.  Reddit: wallstreetbets and Gamestop stock.
+
+```{prf:definition}
+Agents that wait for changes in stock values and rush to trade with
+liquidity providers' standing orders before those standing orders can be
+canceled. As such, being faster than competitors can be profitable.
+```
+
+In response to snipers, liquidity providers might increase spread to
+decrease sniping opportunities. However, this hurts naive agents by
+increasing their effective transaction costs.
+
+## Market Failure Fixes
+
+How can we resolve these market failures?
+
+1.  **Symmetric Speed Bumps:** all orders get delayed by the same
+    amount.
+
+    -   Doesn't solve the speed arms race, nor does it remove sniping
+        risk for liquidity providers.
+
+2.  **Random Speed Bumps:** each order is delayed by some random amount.
+
+    -   Mostly solves the speed arms race (as luck in random difference
+        is more important than speed differences), but makes the sniping
+        risk worse: if a single sniper gets lucky, then the sniping
+        succeeds.
+
+3.  **Sniper-Only Speed Bumps:** only delay snipers.
+
+    -   If implemented, this would resolve the speed arms race and
+        sniping risk, but it is difficult to determine who is a sniper
+        and who is not.
+
+4.  **Frequent Batch Auctions:** batch orders for some short interval
+    and find the market-clearing price at each batch.
+
+    -   Everyone now has time to react to value-changing events as
+        orders are prioritized by price instead of arrival (as long as
+        the order arrives in the same batch).
+
+## Blockchain Flashboys
+
+```{prf:definition}
+Agreement between two agents for one transaction (borrowing) in the
+current period and a second transaction in some future period (paying
+back, perhaps with interest). If the second transaction does not happen,
+some collateral promised by the first agent is given to the second
+agent.
+```
+
+The collateral needs to be sent simultaneously with the first
+transaction (otherwise one agent can abort after getting money).
+
+```{prf:definition}
+A transaction such that either all steps are executed successfully or
+all steps are canceled.
+```
+
+One application of smart contracts is in decentralized finance (DeFi):
+cryptocurrencies, stocks, and contracts are traded. Limit order books
+and automatic market makers can be automated via smart contracts.
+
+```{prf:definition}
+Simultaneously buying low in one market while selling high in another.
+```
+
+Arbitrage is especially attractive with decentralized finance:
+
+1.  Risk-Free Atomic Arbitrage:
+
+    -   In centralized finance, the arbitrageur takes takes risk (if
+        they buy low first, then the high price they wanted to sell at
+        may drop before they get to selling).
+
+    -   However, bundling buy and sell actions into a single atomic
+        transaction resolves this risk.
+
+2.  Cryptocurrencies are highly volatile, leading to more sniping
+    opportunities;
+
+3.  Information (code) is made public so full code and state of smart
+    contract automated market makers are available. As such, it is easy
+    to exploit bugs.
+
+```{prf:definition}
+Transacting right before another agent to take advantage of the
+distortionary effects of the latter's transaction.
+```
+
+In general, this is illegal in centralized finance. In decentralized
+finance however, all transactions are made publicly known before blocks
+are mined. Now, suppose miners are also traders and an arbitrageur finds
+an opportunity. If the arbitrageur broadcasts this transaction, then
+miners can execute the same trade on their own account. Miner includes
+the arbitrageur's transaction on their block but after the miner's own
+transaction.
+
+## Miners' Extractable Value (MEV)
+
+In DeFi, arbitrageurs compete to be at the top of the next block via
+transaction fees. Then, a miner's MEV is how much they can charge an
+arbitrageur. Miners receive rewards for each block they mine and
+additional transaction fees. Originally, the value for mining the block
+itself is greater than the MEV, but in recent years block rewards have
+went down (bitcoin rewards halve every 4ish years) while MEVs have gone
+up (arbitrageurs increasing demand for transactions). When MEVs come
+into play, not all blocks are created equal.
+
+As such, miners can access a new type of fee-sniping attack: if block
+$\hat{b}$ has a larger MEV than block $b$, then miners would prefer to
+mine $\hat{b}$ over $b$. Then, a miner with fraction $\alpha$ of
+computing power profits from a fee-sniping attack if
+
+$$\alpha^2 \cdot \text{higher MEV} > \alpha \cdot \text{lower MEV} \iff \alpha > \frac{\text{lower MEV}}{\text{higher MEV}}.$$
+
+Success probability is even higher if other miners also try to attack
+instead of extending $b_t$.
+
+With selfish tie-breaking, miners want to extend the block with higher
+MEV opposed to the block that was seen first. Furthermore, a miner can
+incentivize other miners to help extend their block by leaving some MEV
+leftover in the next block (known as undercutting). Some possible ways
+to undercut these issues:
+
+1.  Transaction fees distributed among miners that mine the next $k$
+    blocks;
+
+2.  Capped transaction fees;
+
+3.  "Burn" transaction fees instead of giving them to miners.
+
+However, arbitrageurs and miners can always make side payments.