A Junior Quant's Guide to Chasing Vol and Shorting Stocks

700% financing rates never stopped car dealers, so it damn sure won’t stop us.

Outside of capturing vague, esoteric edges, much of quantitative trading boils down to one thing — buying low and selling high.

Take our momentum strategy, for instance:

• Each month, we screen the most actively traded, optionable stocks and rank them using a 12-1 historical lookback combined with a custom momentum metric.

• We then buy the 10 highest-ranked names and rebalance the portfolio monthly.

Here’s a snapshot of this month’s live production performance as of the time of writing:

This was our first month in production, partially fueled by a bit of FOMO after seeing how strong the previous month’s performance was (+40%):

We recommend taking a look at our full breakdown of the strategy, and if you’re really ambitious, give our GitHub repository a look so you can run it yourself:

A Junior Quant's Guide to Time-Series Momentum

Time-Series Momentum GitHub Repository

We’ll be posting another update soon with new insights on execution, basket selection, and more. But for now, we’ve got some more urgent matters to cover.

So, although we’re running an active strategy that’s generating solid returns in a scalable way, we have to confront an uncomfortable truth:

If the broader market goes down, this strategy is dead.

Sure, the basket is relatively uncorrelated, diversified across sectors, and driven by idiosyncratic flows; but in the event of a systemic shock or broad market drawdown, correlations go to one and momentum names tend to fall the fastest and hardest.

At the end of the day, while we are capturing the momentum effect, we’re also essentially just long very high beta. When the market’s strong, we can significantly outperform the S&P; but when things turn south, drawdowns of 50% or more are well within reason.

Once we acknowledge this existential risk, the natural impulse of any trader is to start designing some complex hedging structure — a big protective payout if things go south.

But we’re not going down that road.

Hedging is generally negative expected value (EV) and option prices are pretty efficient, so we’ll often be accepting perpetual underperformance in the good times for only marginal relief in the bad.

Instead, we’d rather take a page from the multi-strategy legends: rather than hedge, diversify.

Specifically, by introducing a strategy that operates on the other side of the equation: short selling.

When we started this experiment, we had never placed a single short trade at scale. In short time, we were pulled into a rich and surprisingly profitable world filled with locates, borrow fees, and volatility-chasing.

The short selling game is massively underappreciated and with a bit of effort, far more accessible than you might think.

We won’t keep you waiting any longer; so, without further ado, let’s dive in.

A Primer on Short Gambling

To begin, we were essentially starting off with a totally blank page:

“Okay, we want to build a strategy that revolves around short selling — now what?”

Our first instinct was to just reverse our existing momentum strategy: if we're already buying the winners, why not just short the losers?

After some testing, it became clear there was a major obstacle:

In the world of short selling, borrow fees are everything.

Since this can get confusing quickly, let’s start with a high-level overview of how shorting actually works:

1. You identify a stock to short.
   Let’s say your momentum system flags Stock XYZ as a loser. You’re running the trade through your brokerage account at Interactive Brokers.

2. IBKR checks if shares are available to borrow.
   You can’t short XYZ if there are no shares going around, so IBKR looks across its network of counterparties to locate inventory.

3. Goldman Sachs has the shares.
   Goldman is holding 2 million shares of XYZ in custody for its clients — long-term holders like pensions or index funds.

4. Goldman lends the shares to IBKR.
   Goldman agrees to lend out the shares (charging a borrow fee), and IBKR passes them along to you, with a markup on top.

5. You short the shares.
   With borrow confirmed, IBKR lets you sell the shares into the open market. You’re now short XYZ and your PnL rises if XYZ drops.

6. ***You start paying borrow fees daily.***
   Here’s the kicker: borrow isn’t free. If XYZ is “hard to borrow” (low inventory), your costs might run 20%, 100%, even 900% annualized!

Interestingly, this entire mechanic — lending out long-held shares for short sellers — is a major business, reliably bringing in billions a year for firms like Goldman and Vanguard. Their “edge” is that they’re not taking market risk by buying shares and lending them out, but rather the original holders (pensions, index funds, clients) still keep the shares they own, Goldman just facilitates the borrow.

Nevertheless, the biggest challenge with short selling is the borrow fee that’s charged daily.

You see, if a stock is commonly known to be garbage with no future, you likely won’t be the only person trying to short it. If your ideal holding period is 1 month (like in our long momentum strategy) and the borrow fee is 300% annualized, you’re going to have to pay nearly 25% of your size:

• Daily interest = Annualized rate / 365 days = 300 /365 = ~0.82%

• Monthly interest = Daily interest * 30 days = 0.82 * 30 = ~25%

However, this need not be the end of our journey.

If the borrow fees are on the higher end, like 300% — that amounts to a daily interest charge of ~0.82%. By shortening our trade horizon from 30 days to just 1 day, the breakeven point shifts — instead of needing a 25% move to be profitable, we only need a 1% move to cover the cost of the borrow.

At that point, borrow fees become a much more manageable hurdle.

So, all we have to do is get exposure to the worst stocks that are likeliest to have the largest 1-day volatilities so that we can beat borrowing costs — easy, right?

Maybe not. But we wouldn’t have made it this far in this business without the willingness to at least try.

Just Predict Vol, Bro. It’s Easy.

Luckily for us, predicting volatility is a lot easier than predicting direction.

So, to begin, let’s re-visit a pretty well-known phenomenon:

Volatility tends to cluster.

In other words, if the past n days were highly volatile, the next n days are likely to be volatile as well.

There are numerous studies on the subject, in fact, it’s one of the earliest hallmark observations in quantitative finance; but you don’t need a research paper to see it in action, just look at any chart of the VIX (implied volatility):

So, if we know that the best predictor of future volatility is recent volatility, we can start by building a simple ranking metric:

1-day historical volatility

In other words, if a stock moved 50% yesterday, can we reasonably expect it to move at least 1% today — enough to cover the borrow fee?

Testing this is relatively trivial, but first we need to define our investable universe.

To avoid survivorship bias, we construct point-in-time universes that reflect which stocks were actively tradable at each moment in history. This way, we won’t foolishly make ourselves look like geniuses for doing things like buying NVDA back when it didn’t even have $1,000 in average daily volume.

Now, unlike our long momentum strategy, we’re not targeting the highest-quality, most liquid names. Instead, we want stocks that are just liquid enough to short — enough for your broker to locate shares, but sketchy enough to avoid just replicating levered beta all over again.

So, we’re going to loosen our criteria from those with the deepest options liquidity to those that had at least $5m in average daily trading volume. This instantly gives us a pool of ~2,500+ eligible stocks to choose from each day.

With this survivorship-bias-free universe in place, we sort by which stocks had the highest 1-day volatility and then observe what volatility looked like the following day.

Here’s a few recent, notable examples:

As you can see, these names aren’t exactly household favorites (KIDZ? GORO?), but this simple heuristic reliably surfaces stocks with high short-term realized volatility.

Now, of course, realized volatility != direction, so although we can reasonably conclude that a stock will have high realized volatility as soon as tomorrow, we’re still in the dark as to whether it’ll go up or down.

In order to make this work, we’re going to need some way of getting an angle on direction. This is the early stage of the research process, so this is the part where you just toss around ideas and see what sticks.

---

Challenge: You have a way of knowing which stocks are going to double overnight, you just need a way to systematically get the direction right, at least 51% of the time.

How will you do it?

Take some time.

---

Well, one idea goes back to what we already know about the SPX-VIX complex — higher implied volatility tends to be associated with lower forward returns:

So, what if we applied that same logic here? If we know a stock is about to experience elevated volatility, why not assume that volatility skews to the downside?

With that idea in mind, we can define a simple strategy:

1. Each day at 3:55 pm (5 mins before market close), we rank all eligible stocks by their realized volatility over the trading day.

2. We then short the top three movers and hold the position overnight.

3. At 3:55 pm the next day, we buy the shares back.

4. Repeat.

Extremely simple, but let’s see how it held up in recent times:

Right away — total bullsh*t. Way too good to be true. Even without including the very real transaction costs, the performance is a bit too good. Over the sample, the 3-stock basket ended up going down ~64% of the time with an average realized volatility of ~13%.

Naturally, we dug deeper, expecting to find some obvious flaw like look-ahead, survivorship bias, or other classic backtesting sins; but surprisingly, there weren’t any.

The logical next step was to simulate accurate borrow fees for each stock on each respective day.

Unfortunately, historical borrow rate data is far messier than your typical OHLCV dataset. Rates vary significantly depending on the lending relationship. For example, Goldman or JPM might already have an agreement with IBKR, offering shares at 65%. After IBKR adds its markup, your cost could end up around 70%.

On the other hand, if you’re routing only through, say, Tastytrade, which might be sourcing borrow from someone like Fidelity / Geode Capital, you could end up paying something like 90% for the exact same ticker.

Rather than spend weeks building a theoretical model around wildly variable and hard-to-source data, we decided to go straight into production.

The goal: find out what the real borrow fees are and whether execution is even feasible.

So, on the first day after our initial test, we ran the daily sort and isolated the optimal candidates:

Next, we needed to figure out how much it would cost us just to be able to short it. So, we pulled up IBKR’s Stock Loan & Borrow tool (SLB) in TWS for each of the names.

Here’s what one looked like:

So, just to be able to short this at all, we would need to pay an annualized rate of nearly 700%! That translates to roughly 2% per day, meaning that the stock would need to drop at least 2% tomorrow to break-even.

Now, remember: our top-3 volatility basket has historically shown average next-day realized volatility of around 13%, so the borrow cost, while high, is far from unfeasible.

Given the high expected movement, we decided to start small. We allocated $1,000 and shorted roughly $300 of each of the three names.

Surprisingly, the orders filled without issue:

Now, to be transparent, for one of the names we did have to play around with the limit orders a bit — the order book was thin, and every time we tried to hit the bid, a new one would appear slightly lower (a dynamic we’ll likely explore later).

Nevertheless, with the trades in place, all that was left was to wait one day to see if our approach was BS or not.

Our expectation wasn’t necessarily to nail the direction on all three, but rather to see outsized volatility — ideally 10% or more.

Eventually, the next day came around and here’s how it went:

For each name, we did a quick search to understand what specifically caused the day’s move; mainly to avoid obvious disqualifiers. One stock, SPTN, had jumped nearly 50% on a takeover announcement. Since prices typically hover near the deal value after such news, we excluded it from consideration due to the likely lack of follow-through volatility.

We then moved forward with shorting the next three highest-ranked names.

While the basket didn’t deliver the upside we were hoping for — returning a slight loss of around 0.51% — several other names in the top 10 did show the kind of explosive volatility we were targeting (i.e., >10%).

As for borrowing costs, we were charged roughly $4 in interest on about $1,000 of total notional exposure; an effective rate of around 0.40%. While still relatively high, not all shares in the basket carried the same borrow rate, so diversifying across names helped reduce costs compared to going all-in on a single ticker.

Encouragingly, when we re-ran the backtest over that specific date, the PnL closely aligned with what we experienced live. That consistency helps validate that our backtest was grounded in some reality, not just noise.

Final Thoughts

We’ve only placed a few live trades to test this, so it’s far too early to draw any definitive conclusions, but nevertheless; there are a few key takeaways.

First, broker logistics are an edge in themselves. While we did have to pay up to access the trades, the fact that we could place them at all already puts us ahead of most retail traders. Nine out of ten brokers wouldn’t have even made the shares available, especially with short locates this obscure.

Second, this reinforces the need for a more refined universe selection process.

We went with a basic, naive top-n approach across thousands of random stocks. It had its moments, but there's significant room for improvement without falling into overfitting. For instance:

• Sector/industry targeting (Pharma/Bio/Therapeutics): These names often experience massive volatility around phase trial results. In many cases, one trial is the company, and if the news is bad, the downside may continue. Pair that with clustered volatility, and the case for shorting looks a lot stronger.

• Event-driven filtering: Sometimes, especially among the “crappier” names, prices drop 20% for no obvious reason (at least not one you'll find with a basic Google search). By filtering for clear idiosyncratic catalysts (analyst downgrades, CEO resignations, cancelled partnerships) we may improve both directional accuracy and follow-through volatility.

• Portfolio sizing: While a 3-stock basket gave us concentrated exposure, expanding to even just the top 10 could drastically improve reliability. The biggest moves won’t always happen in the top 3 — sometimes the #4 or #5 name is the real heavy-hitter.

Now, we aren’t stopping here. While this strategy is still a work in progress (and more of an experiment), it connects directly to another venture we’ve started exploring: replicating our momentum framework in the wild west of memecoins:

Quant Galore Crypto #001: Memecoin Momentum with Ox.Fun

So, even before writing a single line of code, we can already start asking the important questions:

If realized volatility clusters in equities, does it also cluster in memecoins?
If a token drops 50% overnight, are there any residual directional effects the next day?

We’re still early and there’s a lot more testing to do, but this first run shows real potential; and more importantly, it highlights how much edge still exists in the cracks of the market if you're willing to roll up your sleeves and look.

---

Good luck, and happy trading.🫡🫡

---

Still itching for more? Here are a few other deep dives worth your time:

• A Junior Quant's Guide to Generative AI (That Actually Works)

  • Real-world examples of effectively using ChatGPT in quantitative trading — no hallucinatory slop.

• An Inside Look at an Organized Courtsiding Operation

  • Basically the sports version of high-frequency trading but instead of Citadel, it’s guys colluding in Discord groups.

• A Junior Quant's Guide to Ultra-Leveraged Trading

  • If you’re gonna lever up 100-500x your initial capital, here’s how to do it without going broke.