Q: What is the history of the fund?
Launched on August 15, 1997, the Goldman Sachs International Equity Insights Fund uses data-driven models and quantitative techniques to invest in large- and mid-cap companies in developed markets outside of the U.S. and Canada.
The fund is managed by the Quantitative Investment Strategies (QIS) team, which has been a part of the Goldman Sachs Asset Management business since 1989. QIS also manages the Goldman Sachs Small Cap Growth Insights Fund, which shares the same history, philosophy, and process as this fund.
The team invests in equity markets using quantitative techniques to identify attractive investment opportunities. Using data that helps us quantify the quality of a company’s business model, the mispricing in its stock price, the degree of positive sentiment around it, and the themes and trends that it is exposed to, our ultimate aim is to create a portfolio that can outperform the market-cap weighted version of our benchmark, the MSCI EAFE Index. We want the fund’s performance to be consistent and not too volatile while we also seek to ensure the fund is diversified and tracking error is kept at a reasonable level.
Goldman Sachs is a global investment banking, securities, and investment management firm founded in 1869. The firm’s asset management arm, Goldman Sachs Asset Management, L.P., is among the world’s largest asset managers with $1.25 trillion in assets under supervision as of September 30, 2017. Assets under management of QIS are currently about $124 billion; of that, about $31 billion is managed in the family of strategies that includes the International Equity Insights and Small Cap Growth Insights Funds.
Q: How do you define your investment philosophy?
Our philosophy is to remove bias from the investment process by being extremely data driven and disciplined. We identify stocks and evaluate whether they are attractive using a transparent set of criteria. Much of our research focuses on pinpointing specific characteristics that are useful in predicting a company’s future returns, which we then incorporate into our investment model.
Over the last 10 years we’ve been using new technology to quantify previously unquantifiable data, such as sentiment or the latent connections between companies. This machine-run, human-supervised framework includes structured data (numerically based data that can easily be put in spreadsheets) as well as unstructured data (text, pictures, non-numerical data).
Q: What is your investment process?
Because each region that this fund invests in is different than the others, we’ve created separate models tailored to find attractive companies in each of our investable markets: Continental Europe, the U.K., Japan, Australia, and the Far East ex-Japan developed markets of Hong Kong and Singapore. We use a large set of data that is tailored to each region in order to find the best opportunities in each market.
For example, in analyzing sentiment regarding a particular company, we have computers that read things like press releases, earnings call transcripts, and sell-side research reports in multiple languages. This is particularly relevant for an international strategy like ours and ensures that we reach informed and forward-looking conclusions.
To get ahead of the broader market and better position the portfolio, we look for a number of things. For example, we try to identify inflection points. We want to identify whether sentiment is going from good to bad or bad to worse, before the broader market reacts.
Q: How do you discover inflection points?
Rather than running away from the enormous amount of constantly changing information that’s out there, we embrace it with data and technology. One thing we’ve done is define what inflection points might look like in a way that an algorithm or a computer can recognize. This might include programming a computer to read sell-side research in an effort to discern the tone that’s conveyed about companies. More importantly, we measure how analyst tone changes over time and how it compares to the tone being expressed on other peer companies.
For instance, if all stocks in the energy industry suddenly become unattractive, the tone across them all would become more bearish. However, if the tone of one company fell farther than the others, the model would indicate its future returns would be impacted more heavily.
The key output our model produces is a daily stock-by-stock expected return forecast for over 13,000 companies around the world. These views may change every day, ever so slightly; the expected return forecast for a company will fluctuate as our model incorporates new data to reflect the company’s changing fortunes.
Q: Are there any qualitative factors in your strategy?
The qualitative component and the judgment of the portfolio managers is most evident in the model construction itself; we’ve decided what data to analyze, how to weight it, and what to exclude.
Our true value-add comes from how we build, maintain, and continuously enhance that model. From a risk management standpoint more than anything else, there are instances and events where subjective oversight may be warranted. If, for example, the model suggests trades in a direction where we think there are unaccounted risks, we may avoid or minimize those trades.
An example would be leading up to the U.K. referendum in June 2016, Brexit. We measured the portfolio’s sensitivity to a Brexit outcome even though we were agnostic on the referendum outcome. Although our model can point to companies that looked most attractive, and things like sentiment data might have led us in a certain direction, we believed the referendum was too uncertain, and therefore decided to slightly reduce risk in the portfolio, in some respects to immunize our portfolios’ sensitivity to the referendum itself.
Q: How is your model constructed?
Our model has four major components to assess whether or not a stock is attractive. The first, high quality business models, determines whether a company has strong fundamentals and a growing business. It also predicts if its upcoming financial results may surprise on the upside.
Second, a valuation analysis helps us identify mispricing opportunities. Because we are long-term investors, we want to see if the companies we’re looking at are attractively priced relative to their peers.
The third component, sentiment, looks at whether a company has appropriate and improving sentiment in the market. Do people like this company? Is the news and sentiment around it favorable? We even look at data from options and credit default swaps to see if sentiment there is improving.
Examining themes and trends is the model’s fourth element. The focus here is to identify companies that are benefiting from broader price trends in the market and determine if their customers, competitors, suppliers or partners are doing well also.
By putting these four components together, the model identifies the companies most attractive to us: good businesses that are financially secure, that are growing but cheap, that have positive sentiment, and that are exposed to strong themes and trends in the market. There is no event-based data; rather, a set of metrics and characteristics updated every day helps us evaluate the prospects of each investment opportunity
Q: What is your research process?
Across all strategies, we look at over 13,000 companies around the world. Our investment universe for this fund includes the approximately 1,000 stocks in our benchmark, the MSCI EAFE Index.
Our research process begins by creating an expected return forecast for companies. It measures the expected returns relative to a stock’s industry peer group in a particular region – so, for example we’re trying to identify the most and least attractive companies within the European technology sector. This intra-industry, intra-region comparison means we’ll never face a scenario where the most attractive companies out of the 1,000 in the EAFE index are all Japanese utilities.
The next part of the process is based on our philosophy of building a diverse portfolio by investing across various countries and picking the right stocks in the right sectors. Because picking stocks is where the majority of our return comes from – and what the model was created to facilitate – we put guard rails on how much our weights can deviate from the benchmark with respect to sector, industry, and country.
Q: Would you describe your portfolio construction process?
On average, the fund has about 200 stocks, though that number sometimes rises during periods of higher market volatility or falls when volatility is lower.
To maximize excess return while keeping tracking error within our desired range, we use an optimizer to create ideal portfolio weights. Every three days, the optimizer measures each company’s return forecast, risk forecast, and cost forecast. It then decides which companies to overweight and underweight. Stocks that are especially attractive get a high active weight, but if they are risky or costly, that weight may be moderated.
We do have discretion in setting active deviations away from the benchmark, which may change from time to time vs. the benchmark weight. They tend to be in the range of plus or minus 2% – meaning stocks we love may be as high as 2% overweight, while unattractive ones might as much as 2% underweight. Most of our holdings fall in the middle of that range and are typically not near either limit.
Maximum sector and industry positions tend to be about plus or minus 4% the benchmark, but these are revisited from time to time. Country deviations might be larger than 4%, because we feel our country selection is a value-add. However, the portfolio remains one where the majority of active risk comes from our stock selection.
If a stock starts to get risky or its excess return forecast begins to fall, we might decide to replace it with another company in the same industry or sector – keeping the portfolio’s risk profile the same – but will only do so if the cost benefits of the trade also make sense.
Q: How do you define and control risk?
To beat the MSCI EAFE Index, we have to make measured bets against it every time we overweight or underweight a stock, sector, industry, or country – so for us, the most important risk consideration is tracking error, or active risk versus the benchmark. Ours is a modest 200 to 300 basis points on average. It’s such a crucial factor that we don’t think of the portfolio in terms of the number of stocks we own, but instead as the number of bets we’ve made versus the index (stocks that we don’t have a position in, but are benchmark constituents do also add to the portfolio’s tracking error).
We also have our own risk model that focuses on the active risk we take versus the active return we deliver. It’s paramount that we ensure that the ratio of excess return over active risk – or information ratio – is as high as possible. This means we’re getting the greatest amount of excess return while minimizing the excess risk needed to do so.
Q: What lessons did you learn from the financial crisis?
Although 2008 was a landmark moment, 2007 was a period of challenging performance for many quantitative investors. Back then, many of them were using the same types of models, data, and techniques to build portfolios. This worked fine until they all started selling at the same time, which resulted in liquidity-driven derisking and underperformance.
The most important lesson we learned from this is that active managers can best achieve an informational advantage by using unique and alternative data – which is why we’ve fixated so much on these types of data that go into our model.
