Introduction
The momentum factor is the tendency for stocks with stronger past performance to continue to outperform stocks with weaker past performance. While it is like trend-following, which measures a timeseries of an asset’s absolute performance to generate long/ flat/ short signals, cross-sectional momentum ranks assets into relative winners and relative losers. In this post, I’ll focus on replicating the momentum factor in NBA spread markets.
For spread bets, it is common for bettors to compare opponents’ “against-the-spread” (ATS) records for similar contexts (e.g. “Team A is 0-3 ATS as road favorites, while Team B is 3-1 as home underdogs…”). Research suggests that bettors overreact to teams’ relative momentum, such that closing lines create a larger hurdle for the team with more momentum to overcome. While betting closing lines on the higher momentum team is no more profitable than a coin flip, I will investigate whether systematically betting on teams that are better at high momentum stats outperforms betting on teams that are better at low momentum stats. If true, it would support the robustness of the momentum factor to generalize across uncorrelated markets, and identifying which box-score ratios are exhibiting high (low) momentum on a given day enables you to overweight (underweight) inputs to your sports betting process.
Methodology
To translate the momentum factor from the quant finance world to the sports betting world, the Ballin with Beta approach starts by identifying 10 common box-score stats that are known drivers of a team’s performance. Head over to the Factor Playbook for an overview of these 10 stats and an explanation of how they are converted into “funds”.
Each box-score ratio is treated like a fund that bets on every NBA game every day, only betting on the team that has been historically better at the corresponding box-score ratio. Given there are 10 “funds” in the investable universe, every day we would have 10 PnLs, and each of the 10 funds bet on the same games simultaneously. For example, the “offensive field goal percent (FG%) fund” would bet all games on the team with higher historical offensive field goal percentage, while the “offensive assist fund” would bet all games on the team with the higher assist rate on made field goals.
Each day, these 10 “funds” are sorted into quantiles based on their past performance. Just like momentum factor goes long stocks with stronger past performance and short stocks with weaker past performance, each day, the Ballin with Momentum portfolio bets on teams that the 5 “funds” with strongest past performance would bet on and bet against the teams that the 5 “funds” with weakest past performance would bet on, while refraining from betting on both sides of the same game.
Figures 1-3 illustrate the methodology used to construct the long/short momentum replicating portfolio, using only two of the 10 funds and only two of the 8, 4, and 10 games played on 11/12, 11/13, and 11/14/2014, respectively. In this simplified example, the “offensive FG% fund” has stronger performance than the “offensive assist fund” over 11/12 and 11/13, so the Ballin with Momentum portfolio would bet with the “offensive FG% fund” and bet against the “offensive assist fund” on 11/14.
Working Figure 1 from left to right, we start with each team’s trailing 4-game average offensive FG% heading into their 11/12 and 11/13 games. The “offensive FG% fund” splits its 1-unit daily bankroll evenly by betting on the two teams with the higher FG% each day. In this example, it won both bets on 11/12 and lost both bets on 11/13, resulting in a two-day net performance of 0 units.
Likewise, working Figure 2 from left to right, we start with each team’s trailing 4-game average offensive assist rates heading into their 11/12 and 11/13 games. The “offensive assist fund” splits its 1-unit daily bankroll evenly by betting on the two teams with the higher assist rate each day. In this example, it lost both bets on 11/12 and won half of its bets on 11/13, resulting in a two-day net performance of -0.5 units.
Since the “offensive FG% fund’s” past performance of 0 units is higher than that of the “offensive assist fund’s” -0.5 units, on 11/14, the Ballin with Momentum portfolio would bet on all teams that the “offensive FG% fund” would bet on, while betting against all teams that the “offensive assist fund” would bet on. Figure 3 illustrates the vote-based game-weighting approach employed by the Ballin with Momentum portfolio for the 11/14 games, with the long leg voting to bet on Miami and Boston, and the short leg voting against Miami and Boston. In this example, the cancelling votes results in the Ballin with Momentum portfolio refraining from betting at all on 11/14, resulting in a return of 0 units.
While this example uses a simplified two-day lookback to sort box-score ratios into quantiles, I’ll test lookbacks ranging from 3 days to 4 weeks, along with short-term reversal windows ranging from 0 to 7 days. As a refresher on short-term reversal, in the stock market, there is evidence that low volume stocks that exhibited strong one year performance tend to underperform in the next month (short-term). This phenomenon is incorporated in the originally published research on the momentum factor, which ranks stocks’ performance over the last year (excluding the last month) and is something I’ll explore below.
If over the course of multiple seasons, the return of the long/short portfolio vs. closing spread lines is positive and statistically significant across multiple lookback horizons and short-term reversal windows, we can conclude that the momentum factor is robust in sports betting markets.
Results
Using historical box-score and odds data for the 15 NBA regular seasons between 2008 and 2023, Figure 4 summarizes the average daily win percent and corresponding statistical significance \left(t_{Win\ \%}>50\%\right) of the long/short Ballin with Momentum portfolio across multiple lookback horizons and short-term reversal windows. Ignoring short-term reversal, the first row shows that the Ballin with Momentum portfolio had an average daily win percent greater than 50% in 10 out of 11 lookback horizons, with the best win percent of 51.1% landing at the 7-day lookback sweet spot. These findings support the robustness of the momentum factor across lookback horizons, but overlaying a short-term reversal filter meaningfully detracts from performance. Maybe this supports the low volume hypothesis I previously referenced, since all box-score ratio “funds” bet on the same games, which implies they all have the same volume, and makes short-term reversal irrelevant in this framework?
Figure 5 plots the cumulative growth of 1 betting unit invested daily in the long/short momentum portfolio, the long leg (betting on teams the high momentum “winner” funds would bet on), and the short leg (betting against teams the low momentum “loser” funds would bet on), using the 7-day lookback. This analysis assumes no transaction costs (+100 odds on each NBA spread bet).
The Ballin with Momentum long/short portfolio had and an overall average daily win percent of 51.1%, which across 10,452 games over 1,875 betting days in this 15 year span, is statistically greater than 50% at the 90% confidence level \left(t_{Win\ \%\ >\ 50\%}\ =\ 1.79\right) .
Since no bookmaker will give +100 odds on NBA spread bets, this implementation of the momentum factor may not have a large enough risk premium to overcome typical -110 transaction costs by itself; but when combined with other signals, momentum-weighting box-score inputs could enhance a betting strategy’s chances of profitability after paying the bookmakers’ juice.