Special Report
Executive Summary Chart 1Quant Model Prediction Vs. Past Outcomes Complementing the US Political Strategy Quantitative Presidential And Senate Election Models, we introduce our Quantitative House Of Representatives Election Model. Our House election model measures the expected change in seats that will be won or lost by the incumbent party (Democratic Party) in the midterm election. The model predicts that Democrats will lose 21 seats, giving up control of the House and resulting in political gridlock from 2023 to 2025 even if the Democrats somehow hold onto the Senate. The “Blue Sweep” policy setting is effectively over. In a last ditch effort, Democrats will look to pass a budget reconciliation bill before the election. Post-midterm, financial markets will see gridlock as a marginal positive in 2023, as long as inflation levels off. In the very near term, however, US equities still face formidable hurdles that should warrant investors taking a defensive position. Asset Initiation Date Return Long DXY (Dollar Index) 2022-02-23 6.1% Bottom Line: Stay tactically defensive until US election risk subsides and global macro risks stabilize. The 2020 US election was hotly contested and future elections, like the upcoming 2022 midterm election, will be closely watched by investors. BCA’s US Political Strategy has introduced two quantitative models over the past year that aim to predict both the Presidential election in 2024 and the Senate election in 2022. In this report we introduce our House election model, so that we now provide readers with a quantitative model-based estimate for all three major US elections. With the 2022 midterms scheduled for November 8, our House model provides valuable insight into control of Congress in 2023-24. In the 2020 election the Democrats held onto the House while winning the Senate and the White House – the so-called “Blue Sweep.” But the Democrats lost 13 House seats while the GOP gained 14, leaving a mere five-seat margin for President Biden today (221 versus 208 seats today, with six vacancies). In 2022, markets expect Republicans to take control of the House and Senate given the well-established pattern that the president’s party performs badly in midterm elections.1 Our House model agrees, and points to the Democrats losing 21 seats later this year. The Model And Variables Our House model uses a simpler modelling approach than our Presidential and Senate models. Unlike those two models, we do not predict any state level outcomes, nor do we assign a probability to any predictions. For starters, House elections do not occur at the state level but rather at the level of congressional districts. Secondly, we are primarily interested in the overall control of the House rather than individual elections. Therefore our model predicts the number of seats the incumbent party will lose or gain (seat swing), and hence its control of the House. Our model is based off a simple linear regression. Uniquely, in our suite of three models, our House model does not include any economic variables. Rather, the model is based off three independent political variables that explain our dependent variable. Due to data constraints on one of our independent variables, our sample size is limited to 20 observable House elections, from 1982-2020. Our model is defined as: Change In House Seatsi= β0+β1Var1i+β2Var2i+β3Var3i+εi Change In House Seats. This is the dependent variable in the model and what we aim to predict. A negative change means the incumbent party will lose seats while a positive change means the incumbent party will win seats. Congressional approval (Var1). This variable measures the public’s approval rating on “how congress is doing its job.” We take the average net approval rating (approval less disapproval) in an election year. A positive net rating supports the incumbent party in gaining seats while a negative rating does the opposite. Generic congressional ballot (Var2). The generic congressional ballot asks people which party they are likely to vote for in Congress in a given election. We take the average net support rate in an election year (that being whichever party leads the other in congressional ballot polling). The larger the president’s party’s deficit on the generic ballot rate, the more House seats it tends to lose. Defending House seats (Var3). The last independent variable is inspired by work from Sabato’s Crystal Ball.2 This measures the number of House seats defended by the incumbent president’s party in an election year. The more seats to defend, the more seats tend to be lost. One variable we omitted is presidential approval. Readers might find this surprising as presidential job approval ratings have tended to correlate reasonably well with House seats gained and lost in midterm elections. Our reason for excluding this variable is that three explanatory variables explain a high degree of variation in the dependent variable. Combined, our three variables explain more than 80% of the variation in the dependent variable. This is more than satisfactory from a statistical standpoint and keeps the model simplistic in nature. Democrats To Lose The House As it stands, our election model predicts that Democrats will lose control of the House in 2022 (Table 1). The Democrats are predicted to lose 21 seats. This prediction is based off current values of our independent variables as calculated and shown below. For the number of defending House seats, we allocate two of three vacant seats to the Democrats to defend.3 This adds up to 224 seats. Table 1Quant Model Predicts A Democrat Loss For this report, we are only concerned with election outcomes pertaining to midterm elections. In this regard, our model’s prediction is in line with historical outcomes for the president’s party (Chart 1). That is, the president’s party almost always loses House seats in a midterm election. While our model does not provide any probability measure for the predicted outcome, it is in line with market expectations that Democrats will lose the House later in the year. Currently, market implied odds for the Democrats to retain the House are just 16%, as opposed to 87% for Republicans to gain control (Chart 2). Most other private forecasts for the House also point to Democrats losing control.4 Chart 1Quant Model Prediction Vs. Past Outcomes Chart 2Republicans Overwhelmingly Favored To Take The House Back Testing Our Model Chart 3In-Sample Back Testing Results Our House model performs well during in-sample back testing. For in-sample testing, we test our model over our entire sample period (1982-2020) but show results for midterm election outcomes only. Our model correctly predicted the direction of seat change (positive or negative) for 80% of outcomes, missing the direction of seat swing for just the 1998 and 2002 midterm elections (Chart 3). The latter two elections are the only two in the post-WWII period in which the president’s party gained seats and this was due to exceptional circumstances (i.e. the Dotcom Bubble and the September 11, 2001 terrorist attacks). During this same test, when our model correctly predicted the directional change in seat swing, it only over-predicted the change once (in 1986), highlighting a more conservative forecast over time. In 2022, given the stagflationary economic backdrop and President Biden’s weak approval rating, the voting public may very well punish the Democrats harder in November than our model expects. Chart 4Out-Sample Back Testing Results During out-sample back testing, we look at a sample period of 2002-2018, comprising of just five midterm elections. Our model correctly predicts the direction of seat swing in 80% of the midterm elections, just like our in-sample testing showed (Chart 4). 2002 is again a standout election where our model incorrectly predicted the direction of seat swing. Closing In On Election Day The midterm election is approximately five months away. Our Senate model predicts the Democrats will lose control of the Senate. Our House model suggests Democrats will lose the lower chamber too. This view is in line with the consensus across markets, forecasters, and historical outcomes. Given the poor showing by Democrats in the 2020 House election, this House prediction will be hard to change. The Senate race could still see some surprises, such as via the Supreme Court. But all in all the “Blue Sweep” of 2020 is already over. The headwinds against the president’s party have gained even more momentum in the context of high inflation, falling consumer confidence, and low real wages. These factors were not measured in our model, but they do form a basis for voting intentions in elections. Coupled with President Biden’s low approval ratings, in general and in specific policy areas like the economy, the Democratic party will need to pull off a political “Hail Mary” to retain the Senate, let alone the House, later this year. Investment Takeaways President Biden and the Democrats may look to the 1934, 1962, 1998, and 2002 elections for proof that the ruling party can perform well in the midterms. But 1998 was a period of nearly unprecedented peace and prosperity, while 2002 came in the wake of a historic attack on the homeland. The 1934 election reinforced a crisis-era government and as such could serve as a model for Biden, but today’s situation is not as dire as the Great Depression. The 1962 analogy is perhaps the best, since Biden, like President Kennedy during the Cuban Missile Crisis, could conceivably benefit from an escalating showdown with Russia this fall. But Kennedy’s Democrats still lost a net of four seats in the House that year – and Kennedy’s approval rating was above 60% while Biden’s is barely 40%. COVID-19 was an unprecedented shock that continues to play out across the economic and political environment in the US. But while the Republicans suffered from the pandemic itself, the Democrats now own the stagflationary aftermath. Democratic enthusiasm should revive a bit from now until the election, but it would take a massive shock to reverse the general trend. There is a strong correlation between opinion polling in the beginning of the year and the midterm election results. Facing a shellacking, Democrats will make one last-ditch effort to pass a budget reconciliation bill before the election. Given the energy crisis in Europe, there is potential for Biden’s renewable energy subsidies to be repackaged into a general “energy security” bill that drops the former hostility to fossil fuels. This could be matched with limited tax changes, including the 15% minimum corporate tax rate that Biden negotiated with other countries. Otherwise US fiscal policy will virtually freeze even if the Senate stays in Democratic hands. Taxes will no longer be able to rise from 2023 but spending will not be subject to cuts. Heading into 2024, gridlock will be reinforced by our presidential quant election model’s slightly higher odds of Democrats retaining the White House, which we think are underestimated at present. Hence Biden is lined up to retain veto power even if Democrats squander the House and Senate in 2022, as long as his administration avoids a recession. Financial markets will see gridlock as a marginal positive in 2023, as long as inflation levels off. In the very near term, however, US equities still face formidable hurdles that will keep us on the sidelines. Global growth is wobbly. Global supply chains remain constrained, affecting growth outcomes and adding to elevated price levels. China’s zero-covid policy and the absence of a credible plan for US-China tariff reduction and economic re-engagement continue to weigh on sentiment. Fed rate hikes are still generating uncertainty. The Middle East is unstable and likely to bring additional energy supply disruptions. Lastly, the Russia-Ukraine war has yet to come to a ceasefire and Russia is likely to reduce energy supplies to Europe in retaliation for Germany’s energy ban and NATO enlargement. With this backdrop in mind, we remain tactically defensive. We see the potential for improvement after the US election brings a reduction in policy uncertainty – as long as geopolitical risks and inflation also stabilize. Guy Russell Senior Analyst GuyR@bcaresearch.com Statistical Appendix Some clients may be curious to read through our model’s estimated regression coefficients as well as conditional forecasts given certain levels of our independent variables. These are discussed herein: Regression Coefficients As mentioned earlier, our model is estimated exclusively by political variables. The beta coefficients for the three explanatory variables are shown below alongside their t-statistics and p-values. All three of these variables tested statistically significant at 5% and 10% levels. The regression’s R-squared value is 0.8183, meaning that the explanatory variables help explain 81.83% of the variation in the dependent variable (Table A1). Clients can recreate the model’s prediction by multiplying the current level of each variable as it stands today by that variable’s respective beta coefficient and adding the constant at the end of the equation. Be sure to follow the methodology explained earlier in the text if such an exercise is of interest. Also, conditional forecasts can be created by holding certain parameters constant should clients want to better understand what differing levels of the three explanatory variables may imply for the change in House seats. Table A1Regression Coefficients Conditional Forecasts We will create simple conditional forecasts for two explanatory variables. Let’s start with the net congressional approval variable. Forecasts will be generated with data intervals calculated over the course of President Biden’s first term in office so far. The lowest net congressional approval rating was -67 ppt and the highest was -25 ppt. We will use 10 ppt intervals between -20 ppt an -70 ppt. Shading indicates the current level for the variable input. Table A2Conditional Forecasts In Table A2, we hold our other explanatory variables constant at their prevailing levels while we assume differing levels for the net congressional approval variable. We will apply this method to conditional forecasts using the net generic congressional ballot. Table A2 shows us that a more negative net congressional approval rate suggests the president’s party will lose more House seats. Of course, the Democrats cannot lose a fraction of a seat (only whole seats), but this conditional forecast illustrates the point of the variable’s impact on the overall outcome. Table A3Conditional Forecasts Changing to the net generic congressional ballot, our conditional forecast for values ranging from -2 ppt to + 5 ppt are shown in Table A3. This range shows the high and low of the net generic congressional ballot through President Biden’s first term in office. Shading indicates the current level for the variable input. Like before, the results in Table A3 shows us that a more negative net generic congressional ballot suggests the president’s party will lose more House seats. Again, the Democrats cannot lose a fraction of a seat. But the results in Table A3 show that changes in House seats are more sensitive to changes in the net generic congressional ballot variable compared to the net congressional approval variable. This is due to two reasons: The net generic congressional ballot variable is a strong predictor of House election outcomes. It’s larger beta value indicates it’s high degree of sensitivity for the dependent variable, implying that it is the most important variable in our model, and that changes to it have the largest impact on the modelled outcome, that is predicted changes in House seats. Footnotes 1 See The American Presidency Project. "Seats in Congress Gained/Lost by the President's Party in Mid-Term Elections," October 29, 2018. presidency.ucsb.edu 2 See Kyle Kondik, “The Kinds of Seats that Flip in Midterms,” Sabato’s Crystal Ball, May 11, 2022, UVA Center For Politics, centerforpolitics.org. 3 For our 2022 prediction, we allocate vacant House seats evenly between Democrats and Republicans. For example, if there are six vacant seats, each party will be allocated three seats to defend on top of the House seats that they already occupy. If there are an odd number of vacant seats, for example three, each party will receive one seat added to their count, then the incumbent party will receive the remaining seat. 4 See fivethirtyeight.com, 270towin.com and racetowh.com, among others. Strategic View Open Tactical Positions (0-6 Months) Open Cyclical Recommendations (6-18 Months) Table A2Political Risk Matrix Table A3US Political Capital Index Chart A1Presidential Election Model Chart A2Senate Election Model Table A4APolitical Capital: White House And Congress Table A4BPolitical Capital: Household And Business Sentiment Table A4CPolitical Capital: The Economy And Markets
