Wars are often won at the margin and become a battle of attrition. A soccer match were two teams are evenly matched can battle for 90 minutes in open play, then another 30 minutes, for the whole game to be decided by penalty kicks. While imperfect and debatable whether that is enough to define who was better — it does the job in deciding who was the winner — and history will remember mostly that. Many sports are the same.
Asset management has been enduring years of price wars and is being won literally at the margin, net interest margin (NIM). Costs have come down so far that the remaining prize for those willing to fight is the spread between resting cash and whatever return the asset manager can achieve. Additionally, as the world moves towards passive investing there is much less opportunity to attract money flows by virtue of outperformance. With yields generally becoming depressed, a few basis points of outperformance can be enough to win the customer over.
Conceivably we can reach a world where management fees hit a common floor and the tracking error to the underlying index become so low that customers compare the relative tax efficiency between ETFs’ dividends!
This battle of attrition through cost has a long way to go. Whilst the lowest hanging fruit has already been picked, there is still plenty of fruit on the tree. The bigger asset managers will keep picking everything within reach, while the smaller ones will conjure tricks to survive and thrive. The goal of this article is to point out potential areas where there is still plenty of optimization within asset management — classifying them by amount of discretion the portfolio managers have. Perhaps as the industry automates each step, they start looking at the next, with only the largest and efficient managers making it to the top.
Little or no discretion
Foreign cash management
Many investment strategies require handling of multiple currencies. The classic JPY/USD carry trade is one such example. A more current talking point is emerging markets where the valuations have been less flamboyant than the US and many professionals wonder whether we will see some kind of reversion to historical mean performance.
This activity is required by both passive and active managers. A passive manager offering an emerging market passive ETF will have to return everything to a singular currency. Sometimes the same underlying portfolio will be offered in multiple currencies in multiple countries. For passive funds they can choose to automate with code; or codify with humans. The challenge with the latter is that humans — over time — tend to get more expensive whereas machines tend to get cheaper. Additionally, any exception management with humans tends to include a bit more operational risk as you can’t guarantee deterministic responses each time.
Some concrete examples of this activity:
- A PM buys AUD$ bonds which settle in T+2 days. The fund has no AUD$ cash, but lots of USD$. In a larger firm the FX trader will be a central function. They will execute an FX spot transaction which settles on the same day.
- Same as above, but the AUD$ bond may be a forward settlement date. Imagine a passive fund knows it needs to rebalance in 3 weeks; the rest of the street knows this and will attempt to trade ahead of it. The passive manager can trade ahead of the time to save their clients money — this is one of the sources of [hopefully positive] tracking error for passive ETFs. Now the firm has a little discretion: option a) wait — and trade an FX spot trade later as settlement date approaches; or b) trade an FX forward now.
- Dynamic hedging! Let’s say the AUD$ bond is party of a currency hedged ETF. As the AUD$ bond changes in value, the hedges need to be adjusted. As more inflows hit the fund, more AUD$ bonds are being traded and again the hedges need to be adjusted.
- The dreaded T+0 settlement. Notably Chinese bonds settle on the same day. I would hazard a guess that most firms have an exception process for managing cashflows given there is higher political risk of capital controls.
- Variation margin. For products requiring margin payments (futures, cleared swaps, etc) variation margin is another activity that happens in the background. Generally there is a cut-off time to make payments to clearing members/exchanges/CCPs which is T+0. Most firms deal with this by leaving a residual amount of cash in each currency (which impacts returns or creates tracking error).
There are more esoteric examples, but the theme is consistent: there is very little discretion in these transactions. The small amount of discretion is normally set at the firm-level so that all PMs operate on the same rules. Once that is done, there is very little need for humans as markets are already liquid and electronified. Costs can be compressed by more and more automation.
Whilst T+0 remains the outlier there is significant effort and desire to shorten the T+2 that you typically see. http://www.dtcc.com/dtcc-connection/articles/2018/june/22/is-the-industry-ready-for-t-0-settlement. If I were starting a new asset management firm today I would be aiming for intraday sweeps to manage FX transactions in totality — or even event-based FX cash management (more on that below). While moving to that T+0 model isn’t guaranteed but perhaps this is where consumer-facing apps may create some pressure. Creating an app that allows users to send money instantly between two countries has to work around all this plumbing — so as the world becomes more glocal, the need for T+0 is likely to become more accentuated.
The first benefit is operational. For a follow-the-sun organization they need an FX trader for each location that manages sweeps for each regional close. This means that these kind of operational activities define the working day of the organization and become very rigid. Organizations develop ‘blackout’ hours where no core investment activities can occur because the risk of issues is too high. Armies of people, underutilized during morning hours, rush around dealing with whatever operational issue there was leading to the close.
As the market becomes more efficient and transactions become more automated, costs will become more volume-based instead of ticket-based. At the point 2 counterparties negotiate their fees on volume instead of tickets a new world emerges. In the examples above the associated FX transactions are mechanical and can be triggered immediately. If a PM buys a AUD$, a contigent order can be placed on technology’s ‘event bus’; once confirmed the machine can produce a FX trade to compliment. This reduces operational risk because the FX trade is transacted while the PMs context on this overall investment activity is still active. This ‘pair’ trade is complete and there is no need to look at the books at the end of the day and check whether the rest of the org did their jobs properly. This is a massive quality of life improvement! I cannot understate this enough. Saving the organization an hour at the end of the day will give them the same feeling as a day’s vacation!
Imagine the scene. It’s 4pm and the PM receives their ‘flash’ report — which highlights their activity for the day. The FX looks off. Was it an operational error? Did the FX desk decide to hedge tomorrow? Was there a technology problem? Should I stick around and get on the phone to check it out — or deal with it tomorrow?
Executing the FX trade in real-time as a pair trade has a massive impact on improving TCA (Transaction Cost Analysis) and increasing transparency of specific trading strategies. Using the AUD$ bond + FX pair trade as an example — when executed together you get a lot of granular information. The AUD$ bond is the parent order, and the FX trade is the child order. The transaction costs for the FX trade can be rolled up easily into the parent order. Subsequently the performance of the AUD$ bond trade can be judged with the FX costs incorporated into it. Where there are daily FX sweeps it is much harder, from a technology perspective, to attribute the sub-components of the FX trade back to the individual granular reasons as to why it occurred. TCA is a fascinating area, and I’m already going long on this article so will leave it there!
Whilst the business reasons for implementing such as scheme may not transform the absolute returns of the fund, over time I believe the reduction of operational costs will translate to a solid marketing win — reducing the management cost of funds. In this day and age investors are picky about paying 10bps costs in fees — reducing this to 9bps is a marginal advantage as attracting inflows based on lower fees can be differentiating.
Other low/no-discretion activities
FX is only part of the puzzle. There are many other activities that happen impacting the overall portfolio which should be automated or built into the foundation of a system. Knowing them all upfront and leaving room to incorporate them is a challenge and I haven’t seen anyone crush it yet.
Repos/T-Bills/short term securities:
When a fund has cash in its account it generally acts as a drag on the overall performance of the portfolio. The PM may not be able to, or may decide the timing isn’t right, to deploy cash on the day it is received. In the meantime this cash is invested in short term securities. As of today the annualized rate on 4w treasury bill yields about 1.56% and the GC repo rate yields about 1.55%. This is much better than zero!
The implementation of a central repo desk is a big decision as it is complex for many reasons. But generally the reason it gets done is because of the concept of winning at the margin. All equal, a manager that implements a repo desk is going to have better returns than a manager that doesn’t — assuming the total AUM is large enough to cover the costs of the desk.
When most systems were built, repo trading looked very different. Even though it is largely manual today there are more platforms that support centrally cleared tri party repo trading which greases the operational wheels. With the coming SFTR regulations which mandate transaction reporting there will be more incentive to deal on a platform instead of bilaterally.
Whilst automation in this space is not low hanging fruit it deserves a mention as a number of years from now the fruit will be hanging lower in the tree! Some things to keep in mind:
- If you have a central repo desk, from a PM perspective there two impacts: a) cash leaving the fund temporarily; b) a security will be delivered to the portfolio and should be included in the client restrictions
- A systematized repo strategy will require top down guidance from the firm will some PM preferences integrated into the process.
Check out TradeWeb and BrokerTec volumes to keep an eye on how quickly the market place is evolving. Here for a slightly dated overview. Page 8 of the ICMA survey has more details. The TL;DR: more tri party trading, more ATS (Automatic Trading Systems) and less voice. The secular trend is to automation.
How would it be automated? That’s a bit of guesswork but for my 2 cents, it would work something like this:
- The risk analytics produces a view showing anticipated cashflow balances over the coming few days (based on concrete transactions).
- Anticipated inflows/redemptions are fed into this view by account management
- The firm sets a formulaic buffer requirement (e.g. to cover changes in variation margin, etc)
- Open orders that take time to execute are overlaid (e.g. corp bonds that may take days, weeks or months to lean into the position)
- The remaining cash is automatically lent out via the tri party repo rate — on a daily basis.
Not straightforward at all but the key is that once you have identified ‘spare’ cash there is little discretion in how to deploy it. Generally that small amount of discretion is handled centrally so a firm-wide automated process can be implemented. Just to reiterate — this would be costly to do today — but once the market has adjusted to a post-SFTR regime and ATS volumes increase this should become more cost-efficient to implement.
A few more low-discretion events :
- CDS default auctions — bonds for physical delivery
- In-the-money options expiry — cash to buy the underlying; or pair with a trade in the underlying to pocket the difference
- Raising cash for settlement — if there is no cash available a waterfall can be implemented to identify securities to sell to fund the settlement.
- Securities Lending
- A static hedge to be rolled over (E.g. futures rolls, swaps rolls, etc)
I’m sure there are many others but a technology infrastructure needs to model the following to expand to support is:
- Parent/child orders/trades — to model the root trade (i.e. the source of risk/return) and related trades (i.e. hedges, contingent operations, etc)
- Ability to monitor changes in the parent/child trade hierarchy and allow systems to interfere with the subsequent trading decisions/processes
- A robust permissions, logging and event tracing framework to sit on top this to ensure any automation is done in a controlled manner.
- An open architecture that allows for manual intervention via simple APIs (E.g. REST). This allows for programmatic remediation of exceptions instead of copying/pasting between UIs and productivity tools like Excel.
Medium level discretion activities
There is no obvious place to draw the line between low discretion and medium discretion, but I’ll attempt to draw it at the place where the activities can fundamentally impact the overall risk structure of a portfolio. This correlates to the point at which portfolio managers (who manage money across many asset classes) will care at a deeper level about the details.
If a fund wants to gain exposure to a corporate credit spread, but not the risk-free rate (i.e. it likes the corporate but is neutral on the overall risk free term structure), they have a few options to trade:
- Buy the corporate debt and sell some of the treasury holdings from the portfolio
- Buy the corporate debt and sell treasury futures / IR futures
- Buy the corporate debt and trade an IR swap (probably too complicated for most smaller shops)
- Sell CDS protection against the name (I’d classify this more as trading than investment)
Assuming the firm has a centralized rates trader, the PM will typically execute the corp debt trade and then issue an order for the linked rates trade. The notional/quantity on the rates trade is known up upfront and won’t change for as long as the corp debt is held in the portfolio. The rates trade can be executed programmatically quite easily. All we need is the portfolio manager preference on how to execute.
It’s worthy a mention but dynamic hedging gets quite complicated. Assuming the plumbing for a technology solution allows for custom event handlers to be added there is scope for dynamic hedging to be programatic in the future, but in the near term is likely to be the domain of algorithmic/systematic portfolio managers. Here are a few examples — we won’t go into details — maybe in a year or two!
- Currency hedged ETFs. Hedge off the overall portfolio for specific points in time
- Trading options prices — hedging off the underlying market risk buying/selling the underlings. Complicated because of the gamma involved and the potential for prices to change dramatically.
- Evolving leverage — E.g. as prices go higher increase leverage, and prices turn reduce leverage.
Incremental return strategies
There are a number of ways to incrementally improve returns with a decent risk profile. Some of these could be automated.
Covered call writing:
A prudent investor sets their exit price — the price at which they decide their thesis was wrong. You can simply sell when the price hits this level, but a better return is to sell a call option at your exit price.
At least in the case of equities, this is all automatable already. However to implement this across an organization, each PM will need to set the exit price of each investment, give the machine approval to trade options on its behalf. The firm or PM would need to determine the term of the options they would be selling (e.g. nearest option date and continue to rollover; continue to rollover while market price is > x away from the exit price; etc). This needs to be built in combination with a cash management system so that margin is handled automatically.
Carry trades are often described as picking up pennies in front of the steamroller, yet many investment firms like it as a way to boost returns. This article has an example using single currency interest rate swaps and has become more popular as it seems the steamroller is currently out of service. The decision to enter into this trade may take many hard hours of research and soul searching but once decided, the execution is not that tough.
The complexity of this strategy, from a tech perspective, is similar to writing covered calls. The PM needs to decide how much capital to allocate to this strategy, define the mechanics of the strategy, and then have a point at which they want to step out of the trade.
High discretion trading
Relative value opportunities
I put this in the opportunistic side of investments. There are all kinds of analysts in the world trying to spot opportunity and the analysis can come in variety of forms. For a portfolio manger this often comes as recommendations relative to the current portfolio.
For example “I think you should sell your GM stock and buy Ford stock. From our vantage point the latter offers a higher dividend yield, and we don’t see either having any big price appreciation in the coming two years”.
Or “The municipality you own is experiencing a deteriorating demographic trend, we think you should swap for this other municipality where demographics are supportive as you’ll be lowering your default risk”.
Even where you have systematic portfolio management, and systematic relative value models there are a number of challenges to hurdle:
- How do you decide which recommendations to flow into the portfolio?
- You need to check compliance for these trades before trading them
- How much capital do you allow to be influenced by the relative value models?
- How do you attribute performance to the PM who is doing the asset allocation, vs. the quant behind the model who is generating trading recommendations?
- How do you balance additional prospective returns vs. prospective tax liabilities?
Almost a flip side of relative value trading. Imagine the firm’s belief is that active management offers no differentiation of performance. Instead, the PM believes that it’s asset allocation that makes the difference. Perhaps right now, EM looks cheap so we should go overweight, and in exchange go underweight large cap tech growth stocks from the US.
Beyond that the PM may be happy just automating everything downstream from that decision as long as the investment vehicles are cheap (e.g. low cost ETFs).
Many many variations
The asset allocation approach is likely to be easier to automate than the relative value trading approach, but large funds may commingle many different strategies. For example a 2050 target date fund may be comprised of a portion for low cost equity ETFs; a portion for actively managed fixed income and a portion for high risk/high return quant strategies (or a countercyclical component!).
I guess my point is, there is a long way to go to get to extreme automation but the journey starts today and there are many on the road already.
“Full” discretion trading (i.e. AI trading)
There is some hype on full discretion AI type trading strategies but here are a few [subjective!] observations from my side:
- Integrating constraints like customer preferences (e.g. no guns, no tobacco, no oil) into AI strategies is difficult. You could arrive at an ‘optimum’ portfolio ignoring these constraints and post-process, but that post-processing will be programmatic and may detract from the model’s ‘optimum’. Creating and training a model for every individual user is beyond the ability of the industry currently, though we could write models per client persona (‘low risk individual’ vs. ‘max return individual’, vs. ‘eco-investor’, etc).
- Model-ability. There are so many factors that go into a model and every model so far has broken spectacularly at some point once it became popular (portfolio insurance, CDOs, .com per click valuations, etc). For example, can we expect an AI to correlate the corona virus to stock returns? How would it learn this? By learning the impact of SARS and doing a relative impact analysis? If a new regulation is announced — how would this be modelled? Trump’s tweets? Fed announcements? CNBC headlines? Generally the models I have seen be successful end up being some [hyped] variation of a momentum model.
- Perhaps it could be a tree of models. A model for micro-valuation analysis; a model for asset-allocation analysis; a model to dictate portfolio implementation for investor types (e.g. retiree looking for laddered income vs. Harvard endowment fund). But how are the models stitched together? By a human? If a human, then is it still AI?
While I think AI has a long way to go and do see it getting integrated, I think for the next 5–10 years it will remain slightly niche. For example, I wrote about municipals where I think AI could push the industry to the next level, but even then I expect it would be implemented as a standalone trading strategy (ETF, ETMF, a sleeve within a fund, etc), and would still have a few humans sanity checking (e.g. model suggests buying Paradise, CA debt because the yield looks great — but the trade is watching the forest fires on CNBC….).